feat(voices): redesigned My Voices list, working copy, engine-aware TTS availability

My Voices
- Top bar with search, filter toggle (active-count badge), synth-source drawer, group and table buttons;
  filters/sort in a collapsible labelled panel; one-line sort header with select-all/none checkbox
- Groups chip list is a collapsed accordion; bottom bar is Refresh/Sync/Import plus one Actions (N) menu
  (Edit / Audio / Share / Danger), disabled until voices are checked; touch targets stay 44px
- Copy selected works again (copyText was no longer exported), shows "N voice names copied to clipboard"
  and falls back to a copy dialog when the async clipboard is unavailable

TTS / engines
- Health probe works for /v1 card URLs and, behind a LiteLLM gateway, checks the selected model's own
  upstream so stopped containers are not reported as available
- Unhiding a voice publishes a missing entry to the engine's voices.json (no more silent default-voice fallback)
- Status-bar TTS picker also drives the My Voices and STT-test selectors; Use as TTS refreshes the status bar

Includes the engine connect/disconnect, custom TTS, STT footer and magazine auth work that was already
in the working tree, with their tests.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

CHANGELOG.md

@@ -7,15 +7,24 @@ Follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · versioned wi
 
 ## [Unreleased]
 ### Added
+- **My Voices list header:** a sort header (select-all/none checkbox with a partial-selection state, then sortable Name / Lang / Gender / Speed / Rating with the active column highlighted) sits directly above the voice list, and the filter row starts with a filter icon. The full table header still takes over in table view.
 - **QA: Python tests in a throwaway container** (`npm run qa -- --only pyunit --container`): runs the unit tests on production's Python 3.11 in a no-network, read-only container built from the app image on a writable throwaway copy of the read-only mounted repo (`docker/Dockerfile.test` adds only `httpx2` and `nodejs`); opt-in, SKIPPED when docker or the image is missing.
 - **Sonar triage `--drain`:** works through open agent issues one at a time, fires the multica webhook, waits for the agent's answer on GitHub and resends once if nothing arrives; stops on a stalled issue.
 - **Sonar triage can fire a multica webhook autopilot** after each GitHub issue (URL from `~/.config/multica-webhook` or `MULTICA_WEBHOOK_URL`, passed to curl on stdin only); the multica CLI stays as fallback.
 ### Changed
+- **My Voices bottom bar simplified:** Refresh, Sync and Import bundle stay as icon buttons (labels as tooltips / for screen readers) so the whole bar fits one row; every selection-based action now lives in one **Actions (N)** menu grouped Edit / Audio / Share / Danger (disabled until voices are checked; closes on pick, outside click or Esc).
+- **My Voices preview text / synth source** ("Synth uses" Preview text / Reference transcript + sample sentence) moved from its own always-visible card into a drawer behind an icon in the list's top bar (state remembered; the icon tooltip shows the active mode). In the stacked/mobile layout the voice list now has a bounded height and the page scrolls, so opening both drawers no longer squeezes the rows or clips the bottom toolbar.
+- **My Voices filter area redesigned:** one search row with a Filters button (badge = number of active filters) plus group-by-tag / table-view buttons. Language, gender, type, tag, rating, group, sort field/order, "Show disabled", "Clear filters" and the Groups accordion moved into a labelled, collapsible panel (remembered per browser); the sort header above the list stays on one line.
+- **My Voices list:** the Groups chip list is a collapsed accordion (summary shows the group count and the active group, with "Show all"); the bulk actions (Set tag, Set source, Hide, Unhide, Rate, Delete) moved from the selection bar into the bottom toolbar; every selection-based button (Export selected, Copy selected, Precompute, Calc dB, Benchmark, Batch seeds, Verify, Set tag, Set source, Hide, Unhide, Rate, Delete) is disabled, with a tooltip, until voices are checked; only Refresh, Sync and Import bundle are always enabled; the redundant All / None buttons are gone (use the header checkbox).
 - QA: the Sonar triage script groups medium/large findings per rule instead of per (rule, file): one GitHub issue per rule with a file/line table (max 40 rows, 'and N more'; split into 'part n/m' beyond 40 locations); dedupe marker still carries all Sonar keys.
 - Conversation Playground: the round microphone button is red (pressed, "Stop microphone") while the mic is hot (live or recording) and blue ("Start microphone") when off; it returns to blue on stop, auto-stop and mic errors. The pulse ring only animates when reduced motion is not requested.
 - Conversation Playground: the metrics (Token Usage and STT / LLM first token / TTS first audio TTFB charts) now sit in the right-hand Latency column below Turn history instead of a second "Metrics" tab; the Conversation | Metrics tablist is removed. Charts gain axis titles ("Time (ms)", "Time"), a legend and three y-ticks.
 
 ### Fixed
+- **Status-bar TTS picker did not drive My Voices "Play":** choosing an engine in the status bar updated the preview/reader/rehearser selectors but not the My Voices (library) one, so Play kept using the previously selected engine (e.g. a stopped VoiceClone) and failed with a 500. It now also sets the library and Speech-to-Text-test selectors, and the toast names the engine.
+- **Unhidden voices fell back to the default voice:** unhiding moved the files but never registered the voice in the TTS engine's `voices.json` (only saving a transcript or importing did), so the engine logged "Voice … not configured; falling back to default voice" and spoke with another voice. Unhiding now publishes a missing entry from the voice's reference transcript; existing entries (and their cached speaker embeddings) are left alone.
+- **"No TTS backend" although engines show Connected:** the TTS health probe appended `/health` to the card URL (`…:4000/v1/health` = 404), so every custom TTS engine behind a LiteLLM gateway was judged unreachable. Card URLs ending in `/v1` are now probed from the server root, and for engines behind a LiteLLM gateway the probe asks the gateway (`/model/info`, cached 30 s) for the selected model's upstream and checks that engine itself, so a stopped VoiceClone container is shown as unavailable even though the gateway is up (the gateway's unauthenticated `/health/liveliness` is only the fallback); and "Use as TTS / STT" refreshes the status bar immediately.
+- **Copy buttons did nothing** (My Voices "Copy selected", Connect "Copy active voices", engine/voice-inspector/voice-source copy): `copyText` was no longer exported from utils.js after the module scoping, so the click threw `ReferenceError`. It is exported again, now reports whether the copy succeeded, and "Copy selected" shows a copy icon, a "N voice names copied to clipboard" message, and a dialog with the pre-selected list when the async clipboard API is unavailable or refused (plain-HTTP LAN origin, embedded webviews), since the legacy fallback cannot be verified.
 - Sonar: removed 43 small Sonar findings in 20 files: unused variables and dead helper functions (frontend), unused locals and a misread comment (backend); behaviour unchanged. 7 further small findings were deliberately left (used via globals, or entangled with other dead code).
 - Conversation Playground: Turn history rows (#n, time, error rows) were in the DOM but squeezed to zero height by the flex column once charts filled the Latency column; the list no longer shrinks.
 - Conversation Playground: the metrics charts render at the column width (re-rendered on resize) so axis, tick and legend text is 11px instead of ~8px; a single point is centred with a larger dot.

core/config.py

@@ -44,12 +44,12 @@ _SETTINGS_KEYS = {
     # Magazine layout is a vision task. Kept separate from the general text
     # LLM so page images never accidentally go to a text-only model.
     "magazine_layout_url", "magazine_layout_model",
-    "vision_ocr_url", "vision_ocr_model", "vision_ocr_enabled",
+    "vision_ocr_url", "vision_ocr_model", "vision_ocr_enabled", "vision_ocr_api_key", "vision_ocr_fallback_api_key",
     "inline_tags_enabled", "conversation_thinking",
     "vision_ocr_fallback_url", "vision_ocr_fallback_model",
     # OPDS catalogues to search for magazine issues: [{"name": ..., "url": ...}], no credentials.
     "magazine_opds_catalogs",
-    "engine_local_urls", "engine_container_names", "engine_api_keys", "custom_engine_cards",
+    "engine_local_urls", "engine_container_names", "engine_api_keys", "engine_models", "stt_models", "tts_models", "custom_engine_cards", "engine_disabled", "engine_active",
     # Character portrait generation (cloud APIs + local ComfyUI)
     "image_gen_provider", "image_gen_model",
     "comfyui_url", "comfyui_workflow", "comfyui_prompt_node_id",
@@ -140,6 +140,8 @@ def _clean_tts_param_dict(raw, fallback: dict | None = None) -> dict:
 
 
 def _clean_preview_backend(value: str) -> str:
+    if str(value or "").startswith("custom:"):
+        return str(value)
     key = re.sub(r"[^a-z0-9]+", "_", str(value or "voice_clone").lower()).strip("_")
     aliases = {
         "clone": "voice_clone",
@@ -222,6 +224,8 @@ def _strip_tts_extra_params(payload: dict) -> dict:
 
 
 def _post_tts_with_fallback(endpoint: str, payload: dict, headers: dict, **kwargs) -> requests.Response:
+    from core.provider_health import provider_request_params
+    payload = provider_request_params(endpoint, payload)
     resp = request_with_retries("POST", endpoint, json=payload, headers=headers, **kwargs)
     fallback = _strip_tts_extra_params(payload)
     if resp.status_code in {400, 404, 415, 422} and fallback != payload:
@@ -235,7 +239,44 @@ def _post_tts_with_fallback(endpoint: str, payload: dict, headers: dict, **kwarg
     return resp
 
 
+def _engine_card_api_key(settings: dict, name: str) -> str:
+    """Resolve stable IDs and credentials saved under their own legacy name."""
+    keys = settings.get("engine_api_keys") or {}
+    if name in keys:
+        return str(keys[name] or "").strip()
+    card = next((card for card in settings.get("custom_engine_cards", [])
+                 if str(card.get("id") or card.get("name")) == name), None)
+    return str(keys.get(card.get("name")) or "").strip() if card else ""
+
+
+def _tts_api_key(settings: dict, url: str, name: str = "") -> str:
+    """Keep card credentials first and global credentials on their endpoint."""
+    from core.provider_health import provider_key_for_url
+
+    card_key = _engine_card_api_key(settings, name) if name else ""
+    if card_key:
+        return card_key
+    normalized = str(url or "").strip().rstrip("/").removesuffix("/v1")
+    fallback = ""
+    for prefix in ("tts", "llm"):
+        configured = str(settings.get(f"{prefix}_url") or "").strip().rstrip("/").removesuffix("/v1")
+        if normalized and normalized == configured:
+            fallback = str(settings.get(f"{prefix}_api_key") or "").strip()
+            if fallback:
+                break
+    return fallback if name else provider_key_for_url(settings, url, fallback) or fallback
+
+
 def _preview_backend_base_url(settings: dict, backend: str) -> str:
+    if backend.startswith("custom:"):
+        identity = backend.removeprefix("custom:")
+        card = next((card for card in settings.get("custom_engine_cards", [])
+                     if str(card.get("id") or card.get("name")) == identity and card.get("role") == "tts"), None)
+        if not card:
+            return ""
+        urls = settings.get("engine_local_urls") or {}
+        legacy_name = str(card.get("name") or "")
+        return urls.get("dc-" + identity) or urls.get(identity) or urls.get("dc-" + legacy_name) or urls.get(legacy_name) or card.get("url") or ""
     backend = _clean_preview_backend(backend)
     if backend == "streaming":
         return settings.get("tts_stream_url") or settings.get("tts_url") or _TTS_STREAM_DEFAULT
@@ -273,10 +314,17 @@ def _normalize_settings(s: dict) -> dict:
         s["tts_stream_mode"] = "auto"
     if not isinstance(s.get("engine_local_urls"), dict):
         s["engine_local_urls"] = {}
+    if not isinstance(s.get("engine_disabled"), dict):
+        s["engine_disabled"] = {}
+    if not isinstance(s.get("engine_active"), dict):
+        s["engine_active"] = {}
     if not isinstance(s.get("engine_container_names"), dict):
         s["engine_container_names"] = {}
     if not isinstance(s.get("engine_api_keys"), dict):
         s["engine_api_keys"] = {}
+    for key in ("engine_models", "stt_models", "tts_models"):
+        if not isinstance(s.get(key), dict):
+            s[key] = {}
     if not isinstance(s.get("custom_engine_cards"), list):
         s["custom_engine_cards"] = []
     if not isinstance(s.get("magazine_opds_catalogs"), list):
@@ -353,8 +401,15 @@ def _load_settings() -> dict:
         "vision_ocr_fallback_url":   _VISION_OCR_FALLBACK_DEFAULT,
         "vision_ocr_fallback_model": _VISION_OCR_FALLBACK_MODEL_DEFAULT,
         "engine_local_urls":      {},
+        "engine_disabled":        {},
+        "engine_active":          {},
         "engine_container_names": {},
         "engine_api_keys":        {},
+        "engine_models":          {},
+        "stt_models":             {},
+        "tts_models":             {},
+        "vision_ocr_api_key":     "",
+        "vision_ocr_fallback_api_key": "",
         "custom_engine_cards":    [],
         "magazine_opds_catalogs": [],
         "image_gen_provider":     "",

core/magazine.py

@@ -1269,6 +1269,24 @@ def glued_repetition_share(text: str) -> float:
     return covered / len(text)
 
 
+from core.provider_health import provider_request_params
+
+
+def _vision_auth_headers(base_url: str) -> dict[str, str]:
+    """Use only the credential belonging to this configured vision endpoint."""
+    from core.config import _load_settings
+
+    settings = _load_settings()
+    for prefix in ("vision_ocr", "vision_ocr_fallback"):
+        configured = str(settings.get(f"{prefix}_url") or "").strip().rstrip("/")
+        if configured and not urlsplit(configured).path:
+            configured += "/v1"
+        if configured == base_url:
+            key = str(settings.get(f"{prefix}_api_key") or "").strip()
+            return {"Authorization": f"Bearer {key}"} if key else {}
+    return {}
+
+
 def vision_ocr_page(png_path: Path, base_url: str, model: str,
                     timeout: float = 150.0) -> list[dict[str, Any]] | None:
     """OCR one rendered page with a connected vision-capable OCR engine.
@@ -1336,7 +1354,7 @@ def vision_ocr_page(png_path: Path, base_url: str, model: str,
     def _fetch() -> Any:
         resp = request_with_retries(
             "POST", f"{base_url.rstrip('/')}/chat/completions",
-            json=payload, timeout=timeout, attempts=2,
+            json=provider_request_params(base_url, payload, role="ocr"), timeout=timeout, attempts=2, headers=_vision_auth_headers(base_url),
         )
         resp.raise_for_status()
         return resp.json()["choices"][0]["message"]["content"]
@@ -1648,7 +1666,7 @@ def describe_visual_region(image_path: Path, region_type: str, base_url: str,
     def _fetch() -> Any:
         resp = request_with_retries(
             "POST", f"{base_url.rstrip('/')}/chat/completions",
-            json=payload, timeout=timeout, attempts=2,
+            json=provider_request_params(base_url, payload, role="ocr"), timeout=timeout, attempts=2, headers=_vision_auth_headers(base_url),
         )
         resp.raise_for_status()
         return resp.json()["choices"][0]["message"]["content"]

core/provider_health.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 
 from typing import Any, Iterable
+from urllib.parse import urlsplit
 
 import requests
 
@@ -45,3 +46,140 @@ def probe_provider(
         except Exception as exc:
             last_error = str(exc)
     return {'ok': False, 'error': last_error}
+
+
+def provider_key_for_url(settings: dict, url: str, fallback: str = "") -> str:
+    """Resolve saved card credentials for the endpoint actually being called."""
+    normalized = str(url).rstrip("/").removesuffix("/v1")
+    cards = dict(settings.get("engine_local_urls") or {})
+    for card in settings.get("custom_engine_cards") or []:
+        cards[str(card.get("id") or card.get("name") or "")] = card.get("url", "")
+    keys = settings.get("engine_api_keys") or {}
+    for name, card_url in cards.items():
+        if str(card_url).rstrip("/").removesuffix("/v1") == normalized:
+            return str(keys.get(name.removeprefix("dc-")) or keys.get(name) or "")
+    return fallback
+
+
+def _litellm_model_aliases(base: str, key: str) -> list[str]:
+    """Names of a LiteLLM gateway's router aliases (Model Group Alias), which its /models list omits.
+
+    Best effort: the settings endpoint needs an admin-capable key, and other gateways do not have it.
+    """
+    root = base[:-3] if base.endswith("/v1") else base
+    try:
+        r = requests.get(f"{root}/router/settings", timeout=5, headers={"Authorization": f"Bearer {key}"})
+        if r.status_code != 200:
+            return []
+        data = r.json()
+    except Exception:
+        return []
+    if not isinstance(data, dict):
+        return []
+    for holder in (data.get("current_values"), data.get("router_settings"), data):
+        aliases = holder.get("model_group_alias") if isinstance(holder, dict) else None
+        if isinstance(aliases, dict):
+            return [str(name) for name in aliases if name]
+    return []
+
+
+_PROVIDER_REQUIRED_FIELDS = {
+    'model', 'messages', 'input', 'text', 'voice', 'stream', 'response_format',
+    'file', 'files', 'audio', 'audio_prompt', 'audio_prompt_transcript', 'references',
+    'url', 'base_url', 'endpoint', 'headers', 'method',
+}
+_PROVIDER_SECRET_FIELDS = {
+    'apikey', 'authorization', 'proxyauthorization', 'password', 'passwd', 'secret',
+    'token', 'accesstoken', 'refreshtoken', 'privatekey', 'credentials', 'credential',
+    'xapikey', 'accesskey', 'secretkey',
+}
+
+
+def validate_provider_params(params: Any) -> None:
+    """Keep credentials and required request fields out of optional metadata."""
+    if not isinstance(params, dict):
+        raise ValueError('Additional parameters must be a JSON object')
+    if any(key in _PROVIDER_REQUIRED_FIELDS for key in params):
+        raise ValueError('Additional parameters cannot replace required request fields')
+
+    def check(value):
+        if isinstance(value, dict):
+            for key, item in value.items():
+                normalized = ''.join(character for character in str(key).lower() if character.isalnum())
+                if normalized in _PROVIDER_SECRET_FIELDS or normalized == 'headers' or normalized.endswith(('apikey', 'authorization', 'password', 'secret', 'token', 'secretkey', 'privatekey')):
+                    raise ValueError('Credentials must use the API key field')
+                check(item)
+        elif isinstance(value, list):
+            for item in value:
+                check(item)
+    check(params)
+
+
+def engine_disconnected(settings: dict, url: str, role: str = "", identity: str = "") -> bool:
+    """Use explicit selection first; infer legacy endpoints only when unambiguous."""
+    base = str(url).rstrip('/').removesuffix('/v1')
+    urls = dict(settings.get('engine_local_urls') or {})
+    urls.update({str(name).removeprefix('dc-'): value for name, value in list(urls.items()) if str(name).startswith('dc-')})
+    urls = {name: value for name, value in urls.items() if not str(name).startswith('dc-')}
+    from routes.docker import _LOCAL_CONTAINER_DEFS
+    roles = {card['name']: card['role'] for card in _LOCAL_CONTAINER_DEFS}
+    parsed = urlsplit(base)
+    if parsed.hostname in {'localhost', '127.0.0.1', '::1', 'host.docker.internal'}:
+        for card in _LOCAL_CONTAINER_DEFS:
+            if card['port'] == parsed.port:
+                urls.setdefault(card['name'], base)
+    for card in settings.get('custom_engine_cards') or []:
+        name = str(card.get('id') or card.get('name') or '')
+        roles[name] = str(card.get('role') or '')
+        urls.setdefault(name, card.get('url'))
+    if not identity:
+        for setting, selected in (settings.get('engine_active') or {}).items():
+            if str(settings.get(setting) or '').rstrip('/').removesuffix('/v1') == base:
+                if setting == {'llm': 'llm_url', 'ocr': 'vision_ocr_url'}.get(role):
+                    selected_url = urls.get(selected)
+                    if not selected_url or str(selected_url).rstrip('/').removesuffix('/v1') == base:
+                        identity = selected
+                        break
+    if not identity:
+        matches = [name for name, value in urls.items() if (not role or not roles.get(name) or role in roles[name].split('+')) and str(value).rstrip('/').removesuffix('/v1') == base]
+        if len(matches) == 1:
+            identity = matches[0]
+    return (settings.get('engine_disabled') or {}).get(identity) is True
+
+
+def provider_request_params(url: str, payload: dict, settings: dict | None = None, role: str = "", identity: str = "") -> dict:
+    """Merge custom endpoint options while preserving required request fields."""
+    if settings is None:
+        from core.config import _load_settings
+        settings = _load_settings()
+    base = str(url).rstrip('/')
+    routes = {'/audio/transcriptions': 'stt', '/audio/speech': 'tts', '/chat/completions': 'llm', '/transcribe': 'stt', '/tts': 'tts'}
+    for suffix, route_role in routes.items():
+        if base.endswith(suffix):
+            base = base[:-len(suffix)]
+            role = role or route_role
+            break
+    base = base.removesuffix('/v1')
+    if engine_disconnected(settings, base, role, identity):
+        raise ValueError('Engine is disconnected; connect it before use')
+    candidates = []
+    urls = settings.get('engine_local_urls') or {}
+    for card in settings.get('custom_engine_cards') or []:
+        card_role = str(card.get('role') or '')
+        if role and card_role and role not in card_role.split('+'):
+            continue
+        card_identity = str(card.get('id') or card.get('name') or '')
+        if identity and identity != card_identity:
+            continue
+        legacy_name = str(card.get('name') or '')
+        configured = urls.get('dc-' + card_identity) or urls.get(card_identity) or urls.get('dc-' + legacy_name) or urls.get(legacy_name) or card.get('url') or ''
+        if str(configured).rstrip('/').removesuffix('/v1') == base:
+            candidates.append(card)
+    if len(candidates) != 1:
+        return payload
+    params = candidates[0].get('params') or {}
+    try:
+        validate_provider_params(params)
+    except ValueError:
+        return payload
+    return {**payload, **params, **{key: value for key, value in payload.items() if key in _PROVIDER_REQUIRED_FIELDS}}

core/routing.py

@@ -39,6 +39,8 @@ def _clean_route_sound(value: str) -> str:
 
 
 def _clean_route_backend(value: str) -> str:
+    if str(value or "").startswith("custom:"):
+        return str(value)
     key = re.sub(r"[^a-z0-9]+", "_", str(value or "voice_clone").lower()).strip("_")
     aliases = {
         "clone": "voice_clone",

core/tts_helpers.py

@@ -7,11 +7,14 @@ import time
 from pathlib import Path
 
 import requests
+
+from core.provider_health import provider_key_for_url, provider_request_params
 from pydub import AudioSegment
 
 from core.config import (
     _clean_preview_backend,
     _preview_backend_base_url,
+    _tts_api_key,
     _tts_extra_params,
     _apply_tts_extra_params,
     _post_tts_with_fallback,
@@ -119,6 +122,26 @@ def _audio_duration_from_bytes(audio: bytes, media_type: str) -> float | None:
 
 # ── TTS request config & audio ────────────────────────────────────────────────
 
+def _tts_selected_model(settings: dict, keys: str | list[str], default: str) -> str:
+    if isinstance(keys, str):
+        keys = [keys]
+    tts_models = settings.get("tts_models") or {}
+    for key in keys:
+        if model := tts_models.get(key):
+            return str(model)
+    engine_models = settings.get("engine_models") or {}
+    for key in keys:
+        if model := engine_models.get(key):
+            return str(model)
+        if key in engine_models:
+            continue
+        card = next((card for card in settings.get("custom_engine_cards", [])
+                     if str(card.get("id") or card.get("name")) == key), None)
+        if card and (model := engine_models.get(card.get("name"))):
+            return str(model)
+    return default
+
+
 def _tts_request_config(
     text: str,
     voice: str,
@@ -133,17 +156,35 @@ def _tts_request_config(
     from core.constants import _TTS_DEFAULT
     from core.validation import _validate_http_url
 
-    tts_url = _validate_http_url(url_override or settings.get("tts_url", _TTS_DEFAULT), allow_private=True).rstrip("/")
     backend = backend_override or settings.get("tts_backend", "openai")
-    tts_key = (api_key_override if api_key_override is not None else settings.get("tts_api_key", "")).strip()
+    custom_id = backend.removeprefix("custom:") if backend.startswith("custom:") else ""
+    source_url = _preview_backend_base_url(settings, backend) if backend.startswith("custom:") else url_override or settings.get("tts_url", _TTS_DEFAULT)
+    tts_url = _validate_http_url(source_url, allow_private=True).rstrip("/")
+    tts_key = str(api_key_override).strip() if api_key_override is not None else _tts_api_key(settings, tts_url, custom_id)
     tts_hdrs = {"Authorization": f"Bearer {tts_key}"} if tts_key else {}
 
     if backend == "localai":
         endpoint, payload = f"{tts_url}/tts", {"input": text, "model": voice, "response_format": response_format}
     elif backend == "pocket":
-        endpoint, payload = f"{tts_url}/v1/audio/speech", {"input": text, "voice": voice, "response_format": response_format}
+        endpoint, payload = f"{tts_url.removesuffix('/v1')}/v1/audio/speech", {"input": text, "voice": voice, "response_format": response_format}
     else:
-        endpoint, payload = f"{tts_url}/v1/audio/speech", {"model": "tts-1", "input": text, "voice": voice, "response_format": response_format}
+        backend_to_key = {
+            "streaming": "tts_stream_url",
+            "customvoice": "customvoice_url",
+            "nvidia_magpie": "nvidia_tts_url",
+            "kokoro": "kokoro_url",
+            "vibevoice": "vibevoice_url",
+            "xtts": "xtts_url",
+        }
+        if backend.startswith("custom:"):
+            c_id = backend.removeprefix("custom:")
+            model = _tts_selected_model(settings, c_id, "tts-1") # engine_models uses the card ID!
+            endpoint, payload = f"{tts_url.removesuffix('/v1')}/v1/audio/speech", {"model": model, "input": text, "voice": voice, "response_format": response_format}
+        else:
+            bkey = backend_to_key.get(backend, "tts_url")
+            # Fall back to engine_models too if tts_models missed it (we can check both)
+            model = settings.get("tts_models", {}).get(bkey) or _tts_selected_model(settings, [bkey, "tts_url"], "tts-1")
+            endpoint, payload = f"{tts_url.removesuffix('/v1')}/v1/audio/speech", {"model": model, "input": text, "voice": voice, "response_format": response_format}
     if instruct.strip():
         payload["instruct"] = instruct.strip()
     # Include language derived from voice ID prefix so the TTS server knows what
@@ -157,7 +198,7 @@ def _tts_request_config(
     _apply_tts_extra_params(payload, settings, _eff_backend)
     if _eff_backend in ("voice_clone", "streaming", "customvoice"):
         _apply_voice_pinned_seed(payload, voice, settings)
-    return endpoint, payload, tts_hdrs
+    return endpoint, provider_request_params(tts_url, payload, settings, "tts", custom_id), tts_hdrs
 
 
 def _tts_request_audio(
@@ -191,12 +232,16 @@ def _tts_stream_request_config(text: str, voice: str, settings: dict, instruct:
     stream_url = settings.get("tts_stream_url") or settings.get("tts_url") or _TTS_STREAM_DEFAULT
     tts_url = _validate_http_url(stream_url, allow_private=True).rstrip("/")
     tts_key = settings.get("tts_api_key", "").strip()
+    fallback_url = str(settings.get("tts_url") or "").rstrip("/").removesuffix("/v1")
+    if tts_url.rstrip("/").removesuffix("/v1") != fallback_url:
+        tts_key = ""
+    tts_key = provider_key_for_url(settings, tts_url, tts_key)
     tts_hdrs = {"Authorization": f"Bearer {tts_key}"} if tts_key else {}
-    payload = {"model": "tts-1", "input": text, "voice": voice, "response_format": "wav"}
+    payload = {"model": _tts_selected_model(settings, ["tts_stream_url", "tts_url"], "tts-1"), "input": text, "voice": voice, "response_format": "wav"}
     if instruct.strip():
         payload["instruct"] = instruct.strip()
     _apply_tts_extra_params(payload, settings, "streaming")
-    return f"{tts_url}/v1/audio/speech", payload, tts_hdrs
+    return f"{tts_url.removesuffix('/v1')}/v1/audio/speech", provider_request_params(tts_url, payload, settings, "tts"), tts_hdrs
 
 
 def _open_tts_stream_response(text: str, voice: str, settings: dict, instruct: str = "") -> requests.Response:
@@ -265,9 +310,10 @@ def _voice_design_request_audio(
     from core.validation import _validate_http_url
     vd_url = _validate_http_url(settings.get("voice_design_url") or _VOICE_DESIGN_DEFAULT, allow_private=True).rstrip("/")
     vd_key = (settings.get("voice_design_api_key") or settings.get("tts_api_key", "")).strip()
+    vd_key = provider_key_for_url(settings, vd_url, vd_key)
     vd_hdrs = {"Authorization": f"Bearer {vd_key}"} if vd_key else {}
     payload = {
-        "model": _VOICE_DESIGN_MODEL,
+        "model": _tts_selected_model(settings, "voice_design_url", _VOICE_DESIGN_MODEL),
         "input": text,
         "instruct": _apply_voice_design_gender(instruct, gender),
         "language": language,
@@ -293,9 +339,10 @@ def _voice_design_voice_request_audio(
     from core.validation import _validate_http_url
     vd_url = _validate_http_url(settings.get("voice_design_url") or _VOICE_DESIGN_DEFAULT, allow_private=True).rstrip("/")
     vd_key = (settings.get("voice_design_api_key") or settings.get("tts_api_key", "")).strip()
+    vd_key = provider_key_for_url(settings, vd_url, vd_key)
     vd_hdrs = {"Authorization": f"Bearer {vd_key}"} if vd_key else {}
     payload = {
-        "model": _VOICE_DESIGN_MODEL,
+        "model": _tts_selected_model(settings, "voice_design_url", _VOICE_DESIGN_MODEL),
         "input": text,
         "voice": voice,
         "response_format": "wav",
@@ -341,7 +388,7 @@ def _nvidia_clone_request_audio(
     mode = "flow" if str(mode).lower().endswith("flow") else "zeroshot"
     backend = "nvidia_flow" if mode == "flow" else "nvidia_zeroshot"
     base_url = _validate_http_url(_preview_backend_base_url(settings, backend), allow_private=True).rstrip("/")
-    key = (settings.get("voice_design_api_key") or settings.get("tts_api_key", "")).strip()
+    key = provider_key_for_url(settings, base_url)
     headers = {"Authorization": f"Bearer {key}"} if key else {}
     scan_dir = Path(settings.get("voices_scan_dir", _VOICES_DIR_DEFAULT))
     audio_path = _find_voice_audio(voice, scan_dir)
@@ -493,7 +540,9 @@ def _fishspeech_request_audio(
     # room to vary). Every other backend already routes through this same helper; Fish
     # was the one exception.
     _apply_tts_extra_params(payload, settings, "fishspeech")
-    resp = _post_tts_with_fallback(f"{base_url}/v1/tts", payload, {}, timeout=180)
+    key = provider_key_for_url(settings, base_url)
+    headers = {"Authorization": f"Bearer {key}"} if key else {}
+    resp = _post_tts_with_fallback(f"{base_url}/v1/tts", payload, headers, timeout=180)
     resp.raise_for_status()
     audio = resp.content
     if not audio or len(audio) < 256:
@@ -506,7 +555,9 @@ def _fishspeech_request_audio(
 
 def _vibevoice_request_audio(text: str, settings: dict) -> tuple[bytes, str]:
     url = _preview_backend_base_url(settings, "vibevoice").rstrip("/")
-    resp = requests.post(f"{url}/tts", json={"text": text}, timeout=60)
+    key = provider_key_for_url(settings, url)
+    headers = {"Authorization": f"Bearer {key}"} if key else {}
+    resp = requests.post(f"{url}/tts", json=provider_request_params(url, {"text": text}, settings, "tts"), headers=headers, timeout=60)
     resp.raise_for_status()
     audio = resp.content
     if not audio:
@@ -525,6 +576,8 @@ def _preview_request_audio(
     backend: str = "voice_clone",
 ) -> tuple[bytes, str]:
     backend = _clean_preview_backend(backend)
+    if backend.startswith("custom:"):
+        return _tts_request_audio(text, voice, settings, instruct, backend_override=backend, extra_backend=backend)
     if backend == "streaming":
         resp = _open_tts_stream_response(text, voice, settings, instruct)
         audio, media_type = _read_tts_stream_response(resp)
@@ -535,7 +588,7 @@ def _preview_request_audio(
         return _tts_request_audio(
             text, voice, settings, instruct,
             _preview_backend_base_url(settings, "customvoice"),
-            settings.get("tts_api_key", ""),
+            None,
             "openai",
             "customvoice",
         )
@@ -554,7 +607,7 @@ def _preview_request_audio(
         return _tts_request_audio(
             text, voice, settings, instruct,
             _preview_backend_base_url(settings, "nvidia_magpie"),
-            settings.get("tts_api_key", ""),
+            None,
             "nvidia_magpie",
             "nvidia_magpie",
         )
@@ -566,7 +619,7 @@ def _preview_request_audio(
         return _tts_request_audio(
             text, voice, settings, instruct,
             url_override=_preview_backend_base_url(settings, "kokoro"),
-            api_key_override=settings.get("tts_api_key", ""),
+            api_key_override=None,
             backend_override="openai",
             extra_backend="kokoro",
         )
@@ -578,7 +631,7 @@ def _preview_request_audio(
         return _tts_request_audio(
             text, voice, settings, instruct,
             url_override=_preview_backend_base_url(settings, "xtts"),
-            api_key_override=settings.get("tts_api_key", ""),
+            api_key_override=None,
             backend_override="openai",
             extra_backend="xtts",
         )

routes/conversation.py

@@ -398,9 +398,13 @@ def _usage_from_chunk(obj: dict) -> dict | None:
     return out or None
 
 
+from core.provider_health import provider_request_params
+
+
 def _post_llm_stream(url: str, payload: dict, headers: dict):
     """POST a streaming chat request; on 400/422 retry once without the optional keys
     (thinking switch, stream_options) that older backends reject."""
+    payload = provider_request_params(url, payload)
     resp = requests.post(url, json=payload, headers=headers, stream=True, timeout=120)
     optional = [k for k in ("chat_template_kwargs", "think", "stream_options") if k in payload]
     if resp.status_code in (400, 422) and optional:
@@ -429,6 +433,7 @@ def _rewrite_with_persona_sync(text: str, persona: str, llm_url: str, model: str
     if model:
         payload["model"] = model
     payload.update(_thinking_off_params(settings, llm_url))
+    payload = provider_request_params(llm_url, payload, settings, "llm")
     url = f"{llm_url.rstrip('/')}/chat/completions"
     hdrs = {"Authorization": f"Bearer {settings.get('llm_api_key') or 'sk-dummy-key'}"}
     resp = requests.post(url, json=payload, headers=hdrs, timeout=60)
@@ -488,6 +493,7 @@ def _llm_chat_completion_urls(llm_url: str) -> list[str]:
 
 
 def _post_llm_chat_completion(llm_url: str, payload: dict, headers: dict, timeout: float):
+    payload = provider_request_params(llm_url, payload, role="llm")
     last_resp = None
     last_exc = None
     for url in _llm_chat_completion_urls(llm_url):
@@ -745,7 +751,7 @@ async def refine_text(request: Request):
         resp = await asyncio.to_thread(
             requests.post,
             f"{llm_url}/chat/completions",
-            json=payload,
+            json=provider_request_params(f"{llm_url}/chat/completions", payload),
             headers={"Authorization": f"Bearer {settings.get('llm_api_key') or 'sk-dummy-key'}"},
             timeout=60,
         )
@@ -807,7 +813,7 @@ async def rewrite_with_persona(request: Request):
         resp = await asyncio.to_thread(
             requests.post,
             f"{llm_url}/chat/completions",
-            json=payload,
+            json=provider_request_params(f"{llm_url}/chat/completions", payload),
             headers={"Authorization": f"Bearer {settings.get('llm_api_key') or 'sk-dummy-key'}"},
             timeout=600,
         )
@@ -897,7 +903,7 @@ async def analyze_characters(request: Request):
             payload["model"] = model
         resp = requests.post(
             f"{llm_url}/chat/completions",
-            json=payload,
+            json=provider_request_params(f"{llm_url}/chat/completions", payload),
             headers={"Authorization": f"Bearer {_settings.get('llm_api_key') or 'sk-dummy-key'}"},
             timeout=600,
         )
@@ -1045,7 +1051,7 @@ async def match_characters_voices(request: Request):
         try:
             resp = await asyncio.to_thread(
                 requests.post,
-                f"{llm_url}/chat/completions", json=payload,
+                f"{llm_url}/chat/completions", json=provider_request_params(f"{llm_url}/chat/completions", payload),
                 headers={"Authorization": f"Bearer {_settings.get('llm_api_key') or 'sk-dummy-key'}"}, timeout=600,
             )
             resp.raise_for_status()
@@ -1571,7 +1577,7 @@ async def character_sheets_stream(request: Request):
             for u in urls:
                 try:
                     upstream = requests.post(
-                        u, json=payload, headers=headers,
+                        u, json=provider_request_params(u, payload), headers=headers,
                         timeout=(15, prep["timeout_seconds"]), stream=True,
                     )
                     if upstream.status_code in (404, 405) and u != urls[-1]:
@@ -1696,7 +1702,7 @@ async def character_deep_analysis(request: Request):
     try:
         resp = await asyncio.to_thread(
             requests.post,
-            f"{llm_url}/chat/completions", json=payload,
+            f"{llm_url}/chat/completions", json=provider_request_params(f"{llm_url}/chat/completions", payload),
             headers={"Authorization": f"Bearer {_settings.get('llm_api_key') or 'sk-dummy-key'}"}, timeout=600,
         )
         resp.raise_for_status()
@@ -3607,7 +3613,7 @@ async def character_generate_prompts(request: Request):
             payload["model"] = model
         try:
             resp = requests.post(
-                f"{llm_url}/chat/completions", json=payload,
+                f"{llm_url}/chat/completions", json=provider_request_params(f"{llm_url}/chat/completions", payload),
                 headers={"Authorization": f"Bearer {_settings.get('llm_api_key') or 'sk-dummy-key'}"}, timeout=600,
             )
             resp.raise_for_status()
@@ -3806,7 +3812,7 @@ async def audiobook_consistency_check(request: Request):
 
     def _call() -> dict:
         resp = requests.post(
-            f"{llm_url}/chat/completions", json=payload,
+            f"{llm_url}/chat/completions", json=provider_request_params(f"{llm_url}/chat/completions", payload),
             headers={"Authorization": f"Bearer {_settings.get('llm_api_key') or 'sk-dummy-key'}"}, timeout=600,
         )
         resp.raise_for_status()
@@ -5384,7 +5390,7 @@ async def attribute_dialogue_stream(request: Request):
             for u in urls:
                 try:
                     upstream = requests.post(
-                        u, json=payload, headers=headers,
+                        u, json=provider_request_params(u, payload), headers=headers,
                         timeout=(15, prep["timeout_seconds"]), stream=True,
                     )
                     if upstream.status_code in (404, 405) and u != urls[-1]:
@@ -5598,6 +5604,7 @@ async def audio_effects(request: Request):
 _EXPORT_SKIP_KEYS = {
     "groq_api_key", "whisper_api_key", "tts_api_key", "voice_design_api_key",
     "elevenlabs_api_key", "llm_api_key", "engine_api_keys",
+    "vision_ocr_api_key", "vision_ocr_fallback_api_key",
 }
 _IMPORT_ALLOWED_SUFFIXES = set(_AUDIO_EXTS + [".reference.txt", ".meta.json", ".jpg", ".jpeg", ".png", ".webp"])
 
@@ -6058,26 +6065,7 @@ async def mcp_sse(request: Request):
 
 # ── Conversation playground ───────────────────────────────────────────────────
 
-def _litellm_model_aliases(base: str, key: str) -> list[str]:
+from core.provider_health import _litellm_model_aliases
-    """Names of a LiteLLM gateway's router aliases (Model Group Alias), which its /models list omits.
-
-    Best effort: the settings endpoint needs an admin-capable key, and other gateways do not have it.
-    """
-    root = base[:-3] if base.endswith("/v1") else base
-    try:
-        r = requests.get(f"{root}/router/settings", timeout=5, headers={"Authorization": f"Bearer {key}"})
-        if r.status_code != 200:
-            return []
-        data = r.json()
-    except Exception:
-        return []
-    if not isinstance(data, dict):
-        return []
-    for holder in (data.get("current_values"), data.get("router_settings"), data):
-        aliases = holder.get("model_group_alias") if isinstance(holder, dict) else None
-        if isinstance(aliases, dict):
-            return [str(name) for name in aliases if name]
-    return []
 
 
 @router.get("/api/conversation/llm-models")
@@ -6089,6 +6077,9 @@ async def conversation_llm_models(
     from core.validation import _validate_http_url
     settings = _load_settings()
     base = (url or settings.get("llm_url") or "http://localhost:11434/v1").rstrip("/")
+    from core.provider_health import engine_disconnected
+    if engine_disconnected(settings, base, "llm"):
+        return {"models": [], "url": base, "error": "Disconnected", "connected": False}
     key_to_use = provider_api_key or settings.get("llm_api_key") or "sk-dummy-key"
     try:
         base = _validate_http_url(base, allow_private=True)

routes/docker.py

@@ -4,15 +4,15 @@ from __future__ import annotations
 import asyncio
 import os
 from pathlib import Path
-from urllib.parse import quote
+from urllib.parse import quote, urlsplit
 
 import requests
 from fastapi import APIRouter, Header, HTTPException, Request
 
 from core.docker_client import _docker_available, _docker_get_json, _docker_post
-from core.config import _load_settings, _save_settings, _normalize_settings
+from core.config import _load_settings, _save_settings, _normalize_settings, _tts_api_key, _engine_card_api_key
 from core.validation import _normalize_service_url
-from core.provider_health import probe_provider
+from core.provider_health import probe_provider, _litellm_model_aliases
 
 router = APIRouter()
 
@@ -123,6 +123,71 @@ async def restart_local_container(name: str):
     return {"ok": True, "name": name, **_container_status(name)}
 
 
+def _engine_api_key(settings: dict, name: str, role: str, supplied: str = "", url: str = "") -> str:
+    card_key = _engine_card_api_key(settings, name)
+    if role == "tts":
+        return supplied or _tts_api_key(settings, url, name) or "sk-dummy-key"
+    if name:
+        return supplied or card_key or "sk-dummy-key"
+    fallback = {
+        "anthropic": "llm_api_key", "llm": "llm_api_key", "openai": "llm_api_key",
+        "ollama": "llm_api_key", "stt": "whisper_api_key", "tts": "tts_api_key",
+        "ocr": "vision_ocr_api_key",
+    }.get(role, "")
+    key = supplied or card_key or settings.get(fallback, "")
+    return key or "sk-dummy-key"
+
+
+def _engine_headers(key: str, role: str) -> dict:
+    if role == "anthropic":
+        return {"x-api-key": key, "anthropic-version": "2023-06-01"}
+    return {"Authorization": f"Bearer {key}"}
+
+
+@router.post("/api/local-containers/{name}/connection")
+async def engine_connection(name: str, request: Request):
+    body = await request.json()
+    if not isinstance(body, dict) or not isinstance(body.get("connected"), bool):
+        raise HTTPException(400, "connected must be a boolean")
+    settings = _load_settings()
+    known = {card["name"] for card in _LOCAL_CONTAINER_DEFS}
+    known.update(str(card.get("id") or card.get("name")) for card in settings.get("custom_engine_cards", []))
+    if name not in known:
+        raise HTTPException(404, "Unknown engine")
+    active = dict(settings.get("engine_active") or {})
+    urls = dict(settings.get("engine_local_urls") or {})
+    urls.update({str(identity).removeprefix("dc-"): url for identity, url in list(urls.items()) if str(identity).startswith("dc-")})
+    urls = {identity: url for identity, url in urls.items() if not str(identity).startswith("dc-")}
+    roles = {card["name"]: card["role"] for card in _LOCAL_CONTAINER_DEFS}
+    for card in settings.get("custom_engine_cards", []):
+        identity = str(card.get("id") or card.get("name"))
+        urls.setdefault(identity, card.get("url"))
+        roles[identity] = card.get("role", "")
+    def endpoint(url):
+        return str(url or "").rstrip("/").removesuffix("/v1")
+    for role, setting in (("llm", "llm_url"), ("ocr", "vision_ocr_url")):
+        selected_url = endpoint(settings.get(setting))
+        if setting not in active and selected_url:
+            candidates = dict(urls)
+            parsed = urlsplit(selected_url)
+            if parsed.hostname in {"localhost", "127.0.0.1", "::1", "host.docker.internal"}:
+                for card in _LOCAL_CONTAINER_DEFS:
+                    if card["port"] == parsed.port:
+                        candidates.setdefault(card["name"], selected_url)
+            matches = [identity for identity, url in candidates.items() if role in roles.get(identity, "").split("+") and endpoint(url) == selected_url]
+            if matches == [name]:
+                active[setting] = name
+    settings["engine_active"] = active
+    disabled = dict(settings.get("engine_disabled") or {})
+    if body["connected"]:
+        disabled.pop(name, None)
+    else:
+        disabled[name] = True
+    settings["engine_disabled"] = disabled
+    _save_settings(_normalize_settings(settings))
+    return {"ok": True, "connected": body["connected"]}
+
+
 @router.get("/api/probe-url")
 async def probe_url(url: str, type: str = "", name: str = "", provider_api_key: str = Header("", alias="X-Provider-API-Key")):
     """Server-side API probe — checks service-specific endpoints and validates JSON responses."""
@@ -132,30 +197,8 @@ async def probe_url(url: str, type: str = "", name: str = "", provider_api_key:
             base = base[: -len(_suffix)]
             break
     settings = _load_settings()
-    # Key precedence:
+    key_to_use = _engine_api_key(settings, name, type, provider_api_key, base)
-    #  1. A key supplied explicitly by the browser for this probe (e.g. the
+    hdrs = {"User-Agent": "TTS-Voice-Creator/probe", **_engine_headers(key_to_use, type)}
-    #     user just typed/edited it and it isn't saved yet).
-    #  2. The key saved for this specific engine card (settings.engine_api_keys[name]).
-    #     Card values are masked to "__configured__" before ever reaching the
-    #     browser, so on a normal page load the browser has no real secret to
-    #     forward — the backend must resolve it itself from `name`.
-    #  3. The old global llm_api_key, which may belong to a different
-    #     gateway after that card's own key was rotated.
-    card_key = (settings.get("engine_api_keys") or {}).get(name) if name else None
-    key_to_use = provider_api_key or card_key or {
-        "anthropic": settings.get("llm_api_key"),
-        "llm": settings.get("llm_api_key"),
-        "openai": settings.get("llm_api_key"),
-        "stt": settings.get("groq_api_key") or settings.get("whisper_api_key"),
-        "tts": settings.get("tts_api_key"),
-    }.get(type) or "sk-dummy-key"
-    # Anthropic doesn't speak the OpenAI-style Bearer auth every other card
-    # here uses — it needs x-api-key + an anthropic-version header, so the
-    # generic Bearer probe below would 401 even with a perfectly valid key.
-    if type == "anthropic":
-        hdrs = {"User-Agent": "TTS-Voice-Creator/probe", "x-api-key": key_to_use, "anthropic-version": "2023-06-01"}
-    else:
-        hdrs = {"User-Agent": "TTS-Voice-Creator/probe", "Authorization": f"Bearer {key_to_use}"}
 
     if type == "anthropic":
         checks = [("/v1/models", "data")]
@@ -164,7 +207,7 @@ async def probe_url(url: str, type: str = "", name: str = "", provider_api_key:
     elif type == "stt":
         checks = [("/health", None), ("/v1/models", "data"), ("/v1/audio/transcriptions", None)]
     elif type == "tts":
-        checks = [("/health", None), ("/v1/health", None), ("/v1/audio/voices", None), ("/speakers", None), ("/voices", None)]
+        checks = [("/health", None), ("/v1/models", "data"), ("/v1/health", None), ("/v1/audio/voices", None), ("/speakers", None), ("/voices", None)]
     elif type == "ocr":
         checks = [("/health", None), ("/v1/models", "data"), ("/v1/chat/completions", None), ("/parse", None), ("/ocr", None)]
     elif type == "comfyui":
@@ -172,7 +215,7 @@ async def probe_url(url: str, type: str = "", name: str = "", provider_api_key:
     else:
         checks = [("", None)]
 
-    result = probe_provider(base, checks, headers=hdrs, timeout=5)
+    result = await asyncio.to_thread(probe_provider, base, checks, headers=hdrs, timeout=5)
     return {key: value for key, value in result.items() if key != 'payload'}
 
 
@@ -193,12 +236,15 @@ async def use_as_llm(name: str, request: Request, provider_api_key: str = Header
         raise HTTPException(400, "URL is required")
 
     s = _load_settings()
-    card_key = (s.get("engine_api_keys") or {}).get(name)
+    card_key = _engine_card_api_key(s, name)
     key_to_use = provider_api_key or card_key or ""
 
     s["llm_url"] = url
+    s["engine_active"] = {**(s.get("engine_active") or {}), "llm_url": name}
+    s["engine_disabled"] = {key: value for key, value in (s.get("engine_disabled") or {}).items() if key != name}
+    s["llm_api_key"] = key_to_use
+    s["llm_model"] = str(body.get("model") or "").strip()
     if key_to_use:
-        s["llm_api_key"] = key_to_use
         if provider_api_key:
             # Keep this card's own saved key in sync with what was just typed.
             engine_keys = dict(s.get("engine_api_keys") or {})
@@ -209,15 +255,68 @@ async def use_as_llm(name: str, request: Request, provider_api_key: str = Header
     return {"ok": True, "llm_url": url, "key_set": bool(key_to_use)}
 
 
+@router.post("/api/local-containers/{name}/use-as-engine")
+async def use_as_engine(name: str, request: Request, provider_api_key: str = Header("", alias="X-Provider-API-Key")):
+    body = await request.json()
+    if not isinstance(body, dict):
+        raise HTTPException(400, "JSON object is required")
+    role = str(body.get("role") or "")
+    setting_key = str(body.get("setting_key") or "")
+    targets = {
+        "stt": {"whisper_url", "faster_whisper_url", "whisper_cpp_url", "nvidia_asr_url", "nvidia_router_url"},
+        "tts": {"tts_url", "tts_stream_url", "voice_design_url", "customvoice_url", "kokoro_url", "vibevoice_url", "xtts_url", "fishspeech_url", "nvidia_tts_url", "nvidia_clone_url", "nvidia_zeroshot_url", "nvidia_flow_url"},
+        "ocr": {"vision_ocr_url"},
+    }
+    if setting_key not in targets.get(role, set()):
+        raise HTTPException(400, "Unsupported engine setting")
+    url = _normalize_service_url(str(body.get("url") or "")).strip()
+    if not url:
+        raise HTTPException(400, "URL is required")
+    settings = _load_settings()
+    key = provider_api_key or _engine_card_api_key(settings, name)
+    if role == "tts" and not key:
+        key = _tts_api_key(settings, url, name)
+    settings[setting_key] = url
+    settings["engine_active"] = {**(settings.get("engine_active") or {}), setting_key: name}
+    settings["engine_disabled"] = {key: value for key, value in (settings.get("engine_disabled") or {}).items() if key != name}
+    settings["engine_local_urls"] = {**(settings.get("engine_local_urls") or {}), name: url}
+    credential_setting = "voice_design_api_key" if setting_key == "voice_design_url" else {
+        "stt": "whisper_api_key", "tts": "tts_api_key", "ocr": "vision_ocr_api_key",
+    }[role]
+    if setting_key in {"whisper_url", "tts_url", "voice_design_url", "vision_ocr_url"}:
+        settings[credential_setting] = key
+    settings["engine_api_keys"] = {**(settings.get("engine_api_keys") or {}), name: key}
+    model = str(body.get("model") or "").strip()
+    settings["engine_models"] = {**(settings.get("engine_models") or {}), name: model}
+    if role == "ocr":
+        from core.constants import _VISION_OCR_MODEL_DEFAULT
+        settings["vision_ocr_model"] = model or _VISION_OCR_MODEL_DEFAULT
+    elif role == "tts":
+        settings["tts_models"] = {**(settings.get("tts_models") or {}), setting_key: model}
+    else:
+        backend = {"whisper_url": "configured", "faster_whisper_url": "faster_whisper", "whisper_cpp_url": "whisper_cpp", "nvidia_asr_url": "nvidia_parakeet", "nvidia_router_url": "nvidia_router"}[setting_key]
+        settings["stt_models"] = {**(settings.get("stt_models") or {}), backend: model}
+    if body.get("activate") is True:
+        if role == "stt":
+            custom = any(str(card.get("id") or card.get("name")) == name for card in settings.get("custom_engine_cards", []))
+            settings["stt_preferred_backend"] = "custom:" + name if custom else backend
+        elif role == "tts":
+            settings["tts_backend"] = "openai"
+        elif role == "ocr":
+            settings["vision_ocr_enabled"] = True
+    _save_settings(_normalize_settings(settings))
+    return {"ok": True, "key_set": bool(key)}
+
+
 @router.get("/api/engine-models")
-async def engine_models(url: str, type: str = "", provider_api_key: str = Header("", alias="X-Provider-API-Key")):
+async def engine_models(url: str, type: str = "", name: str = "", provider_api_key: str = Header("", alias="X-Provider-API-Key")):
     """Return model IDs from an OpenAI-style engine URL without exposing host-only URLs to the browser."""
     base = _normalize_service_url(url).rstrip("/")
     if not base:
         raise HTTPException(400, "URL is required")
-    candidates: list[str]
+    candidates: list[str] = []
     if base.endswith("/v1"):
-        candidates = [base + "/models", base[:-3].rstrip("/") + "/v1/models"]
+        candidates = [base + "/models", base[:-3].rstrip("/") + "/models"]
     else:
         candidates = [base.rstrip("/") + "/v1/models", base.rstrip("/") + "/models"]
     if type == "ollama":
@@ -226,10 +325,10 @@ async def engine_models(url: str, type: str = "", provider_api_key: str = Header
     seen: set[str] = set()
     models: list[str] = []
     settings = _load_settings()
-    key_to_use = settings.get("llm_api_key") or provider_api_key or "sk-dummy-key"
+    key_to_use = _engine_api_key(settings, name, type, provider_api_key, base)
     for endpoint in candidates:
         try:
-            r = await asyncio.to_thread(requests.get, endpoint, timeout=6, headers={"Authorization": f"Bearer {key_to_use}", "User-Agent": "TTS-Voice-Creator/models"})
+            r = await asyncio.to_thread(requests.get, endpoint, timeout=6, headers={**_engine_headers(key_to_use, type), "User-Agent": "TTS-Voice-Creator/models"})
             if r.status_code >= 400:
                 last_err = f"HTTP {r.status_code} from {endpoint}"
                 continue
@@ -253,6 +352,9 @@ async def engine_models(url: str, type: str = "", provider_api_key: str = Header
                     seen.add(mid)
                     models.append(mid)
             if models:
+                if type in {"llm", "openai", "ollama"}:
+                    aliases = await asyncio.to_thread(_litellm_model_aliases, base, key_to_use)
+                    models = [alias for alias in aliases if alias not in models] + models
                 return {"ok": True, "url": base, "endpoint": endpoint, "models": models}
         except requests.exceptions.Timeout:
             last_err = "Timeout fetching models"

routes/library.py

@@ -460,6 +460,13 @@ async def update_voice_meta(request: Request):
         wav = _move_voice_package(wav, _active_voices_dir(settings) if enabled else _hidden_voices_dir(settings))
         meta = _load_meta(wav)
         meta["enabled"] = enabled
+        if enabled:
+            # Unhiding must make the voice usable: the engine only knows voices listed in its voices.json.
+            try:
+                from routes.tts import publish_voice_clone_if_missing
+                await asyncio.to_thread(publish_voice_clone_if_missing, wav, str(meta.get("lang") or ""), settings)
+            except Exception:
+                pass  # best-effort; the voice is already enabled in our own library
 
     for field in ("note", "rating", "flag", "gender", "age", "lang", "language", "loudness", "persona", "origin", "group",
                   "name", "tag", "source", "avatar", "voice_design_prompt"):

routes/settings.py

@@ -14,6 +14,7 @@ router = APIRouter()
 _SECRET_SETTING_KEYS = {
     "groq_api_key", "whisper_api_key", "tts_api_key", "voice_design_api_key",
     "elevenlabs_api_key", "llm_api_key", "engine_api_keys", "external_api_key",
+    "vision_ocr_api_key", "vision_ocr_fallback_api_key",
 }
 _SECRET_MASK = "__configured__"
 
@@ -55,7 +56,21 @@ async def regenerate_api_key(response: Response):
 @router.post("/api/settings")
 async def post_settings(request: Request):
     data = await request.json()
+    if not isinstance(data, dict):
+        raise HTTPException(400, "JSON object is required")
+    if "custom_engine_cards" in data:
+        from core.provider_health import validate_provider_params
+        cards = data["custom_engine_cards"]
+        if not isinstance(cards, list) or any(not isinstance(card, dict) for card in cards):
+            raise HTTPException(400, "Custom engine cards must be a list of objects")
+        for card in cards:
+            try:
+                validate_provider_params(card.get("params", {}))
+            except ValueError as exc:
+                raise HTTPException(400, str(exc)) from None
     s = _load_settings()
+    previous_cards = {str(card.get("id") or card.get("name") or ""): card for card in s.get("custom_engine_cards", []) if isinstance(card, dict)}
+    previous_maps = {key: dict(s.get(key) or {}) for key in ("engine_api_keys", "engine_models", "engine_local_urls")}
     clear = {str(k) for k in (data.get("clear_secrets") or [])} if isinstance(data.get("clear_secrets"), list) else set()
     for key, value in data.items():
         if key not in _SETTINGS_KEYS:
@@ -76,6 +91,27 @@ async def post_settings(request: Request):
             if value in (None, "") and key not in clear:
                 continue
         s[key] = value
+    if isinstance(data.get("custom_engine_cards"), list):
+        for card in data["custom_engine_cards"]:
+            if not isinstance(card, dict):
+                continue
+            identity = str(card.get("id") or card.get("name") or "")
+            previous = previous_cards.get(identity)
+            if not previous:
+                continue
+            old_name = str(previous.get("name") or "")
+            for key in previous_maps:
+                source = "dc-" + old_name if key == "engine_local_urls" else old_name
+                target = "dc-" + identity if key == "engine_local_urls" else identity
+                values = dict(s.get(key) or {})
+                if key == "engine_api_keys" and (identity in clear or old_name in clear):
+                    continue
+                if target not in values:
+                    if target in previous_maps[key]:
+                        values[target] = previous_maps[key][target]
+                    elif source in previous_maps[key]:
+                        values[target] = values.get(source, previous_maps[key][source])
+                    s[key] = values
     s = _normalize_settings(s)
     _save_settings(s)
     return {"ok": True}

routes/stt.py

@@ -21,7 +21,7 @@ from core.constants import (
 )
 from core.registry import _registry_get, TEMP_DIR
 from core.validation import _validate_http_url, _copy_limited, _validated_upload_suffix
-from core.provider_health import probe_provider
+from core.provider_health import probe_provider, provider_key_for_url, provider_request_params, engine_disconnected
 from core.audio import _to_wav_16k
 from core.database import background_job_create, background_job_get, background_job_update
 from core.voice import _AUDIO_EXTS, _UPLOAD_EXTS
@@ -152,7 +152,9 @@ def _stt_backend_url(settings: dict, backend: str) -> str:
         custom_id = backend[len("custom:"):]
         for card in settings.get("custom_engine_cards", []):
             if str(card.get("id", card.get("name", ""))) == custom_id:
-                return card.get("url", "")
+                urls = settings.get("engine_local_urls") or {}
+                name = str(card.get("name") or "")
+                return urls.get("dc-" + custom_id) or urls.get(custom_id) or urls.get("dc-" + name) or urls.get(name) or card.get("url", "")
         return ""
     if backend == "nvidia_parakeet":
         return settings.get("nvidia_asr_url") or _NVIDIA_ASR_DEFAULT
@@ -183,8 +185,18 @@ def _stt_backend_configured(settings: dict, backend: str) -> bool:
     return any(str(settings.get(key) or '').strip() for key in keys)
 
 
-def _stt_backend_model(backend: str) -> str:
+def _stt_backend_model(backend: str, settings: dict | None = None) -> str:
     backend = _clean_stt_backend(backend)
+    settings = settings if settings is not None else _load_settings()
+    selected = (settings.get("stt_models") or {}).get(backend)
+    if backend.startswith("custom:"):
+        identity = backend[7:]
+        card = next((card for card in settings.get("custom_engine_cards", [])
+                     if str(card.get("id") or card.get("name") or "") == identity), {})
+        models = settings.get("engine_models") or {}
+        selected = models.get(identity, models.get(card.get("name")) or selected)
+    if selected:
+        return str(selected)
     if backend in {"nvidia_parakeet", "nvidia_router"}:
         return "whisper-1"
     if backend == "whisper_cpp":
@@ -228,7 +240,24 @@ def _stt_backend_label(backend: str, url: str) -> str:
 def _stt_backend_api_key(settings: dict, backend: str) -> str:
     if backend == "groq_whisper":
         return settings.get("groq_api_key", "").strip()
-    return settings.get("whisper_api_key", "").strip()
+    if backend.startswith("custom:"):
+        from core.config import _engine_card_api_key
+        return _engine_card_api_key(settings, backend[7:])
+    setting = {"configured": "whisper_url", "faster_whisper": "faster_whisper_url", "whisper_cpp": "whisper_cpp_url", "nvidia_parakeet": "nvidia_asr_url", "nvidia_router": "nvidia_router_url"}.get(backend)
+    selected = (settings.get("engine_active") or {}).get(setting)
+    if selected:
+        from core.config import _engine_card_api_key
+        card_key = _engine_card_api_key(settings, selected)
+        if card_key:
+            return card_key
+        if any(selected == str(card.get("id") or card.get("name") or "") for card in settings.get("custom_engine_cards", [])):
+            return ""
+        url = _stt_backend_url(settings, backend).rstrip("/").removesuffix("/v1")
+        configured_url = str(settings.get("whisper_url") or "").rstrip("/").removesuffix("/v1")
+        return str(settings.get("whisper_api_key") or "").strip() if url == configured_url else ""
+    url = _stt_backend_url(settings, backend)
+    fallback = settings.get("whisper_api_key", "") if url.rstrip("/") == str(settings.get("whisper_url") or "").rstrip("/") else ""
+    return provider_key_for_url(settings, url, fallback).strip()
 
 
 def _make_minimal_wav(duration_ms: int = 500, sample_rate: int = 16000) -> bytes:
@@ -244,7 +273,7 @@ def _make_minimal_wav(duration_ms: int = 500, sample_rate: int = 16000) -> bytes
     return header + data
 
 
-def _stt_backend_health(url: str) -> dict:
+def _stt_backend_health(url: str, api_key: str = "") -> dict:
     base = _validate_http_url(url, allow_private=True).rstrip("/")
     # Users commonly paste an OpenAI-style `/v1` base URL. WhisperX exposes
     # health at the service root, while its transcription route is under
@@ -252,10 +281,11 @@ def _stt_backend_health(url: str) -> dict:
     # `/v1/v1/models`, incorrectly marking a healthy WhisperX service down.
     api_base = base[:-3].rstrip("/") if base.endswith("/v1") else base
     models: list[str] = []
-    health = probe_provider(api_base, [("/health", None)], timeout=2)
+    headers = {"Authorization": f"Bearer {api_key}"} if api_key else {}
+    health = probe_provider(api_base, [("/health", None)], headers=headers, timeout=2)
     models_result = probe_provider(
         api_base, [("/v1/models", "data")],
-        headers={"Authorization": "Bearer sk-dummy-key"}, timeout=3,
+        headers=headers, timeout=3,
     )
     ok = bool(health['ok'] or models_result['ok'])
     payload = models_result.get('payload')
@@ -297,9 +327,11 @@ def _transcription_text_from_response(resp: requests.Response) -> str:
 def _transcribe_audio(src: Path, settings: dict, backend: str = "configured") -> tuple[str, str]:
     backend = _clean_stt_backend(backend)
     stt_url = _validate_http_url(_stt_backend_url(settings, backend), allow_private=True).rstrip("/")
+    if _stt_disconnected(settings, backend, stt_url):
+        raise HTTPException(409, "STT engine is disconnected; connect it before use")
     stt_key = _stt_backend_api_key(settings, backend)
     hdrs = {"Authorization": f"Bearer {stt_key}"} if stt_key else {}
-    model = _stt_backend_model(backend)
+    model = _stt_backend_model(backend, settings)
     # Accept both `http://host:port` and OpenAI-style `http://host:port/v1`.
     # WhisperX's native fallback remains rooted at `/transcribe`.
     if stt_url.endswith("/v1"):
@@ -315,7 +347,7 @@ def _transcribe_audio(src: Path, settings: dict, backend: str = "configured") ->
             resp = requests.post(
                 endpoint,
                 files={"file": ("audio.wav", f, "audio/wav")},
-                data={"model": model, "response_format": "text"},
+                data={key: json.dumps(value) if isinstance(value, (dict, list)) else value for key, value in provider_request_params(stt_url, {"model": model, "response_format": "text"}, settings, "stt", backend[7:] if backend.startswith("custom:") else "").items()},
                 headers=hdrs,
                 timeout=_STT_REQUEST_TIMEOUT,
             )
@@ -328,7 +360,7 @@ def _transcribe_audio(src: Path, settings: dict, backend: str = "configured") ->
             resp = requests.post(
                 f"{stt_url}{path}",
                 files={"file": ("audio.wav", f, "audio/wav")},
-                data={"model": "whisper-1", "response_format": "text"},
+                data={key: json.dumps(value) if isinstance(value, (dict, list)) else value for key, value in provider_request_params(stt_url, {"model": "whisper-1", "response_format": "text"}, settings, "stt", backend[7:] if backend.startswith("custom:") else "").items()},
                 headers=hdrs,
                 timeout=60,
             )
@@ -487,6 +519,43 @@ async def _default_stt_benchmark_engines() -> list[dict]:
 
 # ── STT routes ────────────────────────────────────────────────────────────────
 
+def _stt_backend_engine_id(settings: dict, backend: str, url: str = "") -> str:
+    identities = {"faster_whisper": "faster-whisper", "whisper_cpp": "whisper-cpp", "nvidia_parakeet": "parakeet-stt-multi-9011", "nvidia_router": "nvidia-parakeet-asr"}
+    setting = {"configured": "whisper_url", "faster_whisper": "faster_whisper_url", "whisper_cpp": "whisper_cpp_url", "nvidia_parakeet": "nvidia_asr_url", "nvidia_router": "nvidia_router_url"}.get(backend)
+    if backend.startswith("custom:"):
+        return backend[7:]
+    selected = (settings.get("engine_active") or {}).get(setting)
+    if selected:
+        return selected
+    from routes.docker import _LOCAL_CONTAINER_DEFS
+    for card in _LOCAL_CONTAINER_DEFS:
+        if card["role"] == "stt" and str(card["port"]) == _backend_port_label(url):
+            return card["name"]
+    return identities.get(backend, "")
+
+
+def _stt_disconnected(settings: dict, backend: str, url: str) -> bool:
+    from routes.docker import _LOCAL_CONTAINER_DEFS
+    native_identity = _stt_backend_engine_id({}, backend)
+    identity = _stt_backend_engine_id(settings, backend, url)
+    selected_native = any(card["name"] == identity and card["role"] == "stt" for card in _LOCAL_CONTAINER_DEFS)
+    if not selected_native and (settings.get("engine_disabled") or {}).get(native_identity) is True:
+        return True
+    return engine_disconnected(settings, url, "stt", identity)
+
+
+def _stt_backend_title(settings: dict, backend: str, url: str) -> str:
+    from routes.docker import _LOCAL_CONTAINER_DEFS
+    setting = {"configured": "whisper_url", "faster_whisper": "faster_whisper_url", "whisper_cpp": "whisper_cpp_url", "nvidia_parakeet": "nvidia_asr_url", "nvidia_router": "nvidia_router_url"}.get(backend)
+    selected = (settings.get("engine_active") or {}).get(setting)
+    for card in settings.get("custom_engine_cards", []):
+        if selected and selected == str(card.get("id") or card.get("name") or ""):
+            return card.get("label") or card.get("name") or "Custom STT"
+    for card in _LOCAL_CONTAINER_DEFS:
+        if card["name"] == selected or (not selected and card["role"] == "stt" and str(card["port"]) == _backend_port_label(url)):
+            return card["label"]
+    return "Configured STT" if backend == "configured" else _stt_backend_label(backend, "")
+
 @router.get("/api/stt-backends")
 async def stt_backends():
     settings = _load_settings()
@@ -501,36 +570,47 @@ async def stt_backends():
             continue
         seen_urls.add(key)
         api_key = _stt_backend_api_key(settings, backend)
-        if backend == "groq_whisper":
+        disconnected = _stt_disconnected(settings, backend, url)
+        if disconnected:
+            ok, models, health_detail = False, [], {"ok": False, "error": "Disconnected"}
+        elif backend == "groq_whisper":
             ok = bool(api_key)
+            health_detail = {"ok": ok}
             models: list[str] = ["whisper-large-v3-turbo", "whisper-large-v3", "distil-whisper-large-v3-en"]
         else:
-            health_detail = _stt_backend_health(url)
+            health_detail = _stt_backend_health(url, api_key)
             ok, models = health_detail.get("ok", False), health_detail.get("models", [])
         items.append({
             "id": backend,
+            "engine_id": _stt_backend_engine_id(settings, backend, url),
             "label": _stt_backend_label(backend, url),
+            "title": _stt_backend_title(settings, backend, url),
             "url": url,
             "port": _backend_port_label(url),
             "available": ok,
+            "connected": not disconnected,
             "configured": _stt_backend_configured(settings, backend),
             "health": {key: health_detail.get(key) for key in ("ok", "status", "endpoint", "error") if health_detail.get(key) is not None},
-            "model": _stt_backend_model(backend),
+            "model": _stt_backend_model(backend, settings),
             "models": models,
             "metrics": _STT_BACKEND_METRICS.get(backend, {}),
         })
     for card in settings.get("custom_engine_cards", []):
         if card.get("role") not in ("stt", "stt+tts"):
             continue
-        raw_url = card.get("url", "").strip()
-        url = _validate_http_url(raw_url, allow_private=True).rstrip("/")
-        if not url:
-            continue
         card_id = "custom:" + str(card.get("id", card.get("name", "")))
+        raw_url = _stt_backend_url(settings, card_id).strip()
+        if not raw_url:
+            continue
+        try:
+            url = _validate_http_url(raw_url, allow_private=True).rstrip("/")
+        except HTTPException:
+            continue
         if (card_id, url) in seen_urls:
             continue
         seen_urls.add((card_id, url))
-        health_detail = _stt_backend_health(url)
+        disconnected = _stt_disconnected(settings, card_id, url)
+        health_detail = {"ok": False, "error": "Disconnected"} if disconnected else _stt_backend_health(url, _stt_backend_api_key(settings, card_id))
         ok, models = health_detail.get("ok", False), health_detail.get("models", [])
         port = _backend_port_label(url)
         label = card.get("label") or card.get("name") or "Custom STT"
@@ -538,13 +618,16 @@ async def stt_backends():
             label = f"{port} {label}"
         items.append({
             "id": card_id,
+            "engine_id": card_id[7:],
             "label": label,
+            "title": card.get("label") or card.get("name") or "Custom STT",
             "url": url,
             "port": port,
             "available": ok,
+            "connected": not disconnected,
             "configured": _stt_backend_configured(settings, card_id),
             "health": {key: health_detail.get(key) for key in ("ok", "status", "endpoint", "error") if health_detail.get(key) is not None},
-            "model": "whisper-1",
+            "model": _stt_backend_model(card_id, settings),
             "models": models,
             "metrics": {},
         })

routes/tts.py

@@ -21,7 +21,7 @@ from fastapi import APIRouter, HTTPException, Request
 from fastapi.responses import Response, StreamingResponse
 
 from core.tasks import spawn
-from core.config import _load_settings, _clean_preview_backend, _preview_backend_base_url
+from core.config import _load_settings, _clean_preview_backend, _preview_backend_base_url, _tts_api_key
 from core.constants import (
     _VOICES_DIR_DEFAULT, _TTS_CONTAINER, _TTS_CONTAINERS_RAW,
     _routing_log_add, CONFIG_DIR, _VOICE_PEAK_DBFS,
@@ -42,7 +42,7 @@ from core.voice import (
 from core.voice_index import indexed_voices, refresh_voice_index_background
 from core.validation import _validate_http_url
 from core import ssml_lite
-from core.provider_health import probe_provider
+from core.provider_health import probe_provider, engine_disconnected
 from core.registry import _registry_put, TEMP_DIR
 from core.tts_helpers import (
     _tts_request_audio, _open_tts_stream_response, _read_tts_stream_response,
@@ -214,7 +214,7 @@ def _backend_port_label(url: str) -> str:
     return ""
 
 
-def _backend_display_name(backend: str, url: str) -> str:
+def _backend_display_name(backend: str, url: str, settings: dict = None) -> str:
     names = {
         "voice_clone":    "Voice Clone/Base (WAV File)",
         "voice_design":   "Voice Design",
@@ -228,11 +228,42 @@ def _backend_display_name(backend: str, url: str) -> str:
         "xtts":           "XTTS v2",
         "fishspeech":     "Fish Audio S2 Pro (Clone)",
     }
-    port = _backend_port_label(url)
+
-    return f"{port} {names.get(backend, backend)}" if port else names.get(backend, backend)
+    prefix = ""
+    if settings:
+        from core.tts_helpers import _tts_selected_model
+        bkeys = {
+            "voice_clone": "tts_url",
+            "voice_design": "voice_design_url",
+            "streaming": "tts_stream_url",
+            "customvoice": "customvoice_url",
+            "nvidia_magpie": "nvidia_tts_url",
+            "kokoro": "kokoro_url",
+            "vibevoice": "vibevoice_url",
+            "xtts": "xtts_url",
+            "fishspeech": "fishspeech_url"
+        }
+        if bkey := bkeys.get(backend):
+            model = _tts_selected_model(settings, [bkey, "tts_url"], "")
+            if model and model not in ("tts-1", "tts-1-hd"):
+                prefix = model
+
+    if not prefix:
+        prefix = _backend_port_label(url)
+
+    return f"{prefix} {names.get(backend, backend)}" if prefix else names.get(backend, backend)
 
 
 def _backend_capabilities(backend: str) -> dict:
+    if backend.startswith("custom:"):
+        return {
+            "purpose": "Custom TTS Engine",
+            "identity": "Custom configured engine behavior.",
+            "style": "Behavior depends on the underlying engine.",
+            "best_for": "Using custom externally configured APIs.",
+            "uses_wav": False, "style_aware": False, "true_streaming": False,
+            "speed": "varies", "latency": "varies", "quality": "varies", "ram": "varies",
+        }
     caps = {
         "voice_clone": {
             "purpose": "Clone a speaker from a short WAV/reference clip.",
@@ -336,8 +367,75 @@ def _backend_capabilities(backend: str) -> dict:
     return caps.get(_clean_preview_backend(backend), {})
 
 
-def _backend_health_detail(url: str) -> dict:
+_TTS_BACKEND_CARDS = {
-    return probe_provider(url, [("/health", None), ("/v1/health", None), ("/v1/audio/list_voices", None)], timeout=2)
+    "voice_clone": "tts-faster-qwen3-tts-voiceclone",
+    "voice_design": "tts-faster-qwen3-tts-voicedesign",
+    "customvoice": "tts-faster-qwen3-tts-customvoice",
+    "streaming": "tts-faster-qwen3-tts-streaming",
+    "fishspeech": "fish-speech-api",
+    "nvidia_magpie": "nvidia-magpie-tts",
+    "nvidia_zeroshot": "nvidia-zeroshot-nim",
+    "nvidia_flow": "nvidia-flow-nim",
+    "kokoro": "kokoro-tts",
+    "vibevoice": "vibevoice-msft-tts",
+    "xtts": "xtts-dgx-spark",
+}
+
+
+def _backend_card_title(backend: str) -> str:
+    from routes.docker import _LOCAL_CONTAINER_DEFS
+
+    return next((card["label"] for card in _LOCAL_CONTAINER_DEFS
+                 if card["name"] == _TTS_BACKEND_CARDS.get(backend)), backend)
+
+
+_GATEWAY_UPSTREAMS: dict[str, tuple[float, dict[str, str]]] = {}
+_GATEWAY_UPSTREAM_TTL = 30.0
+
+
+def _gateway_upstream(root: str, api_key: str, model: str) -> str:
+    """api_base of ``model`` on a LiteLLM gateway, or "" when unknown / not a gateway.
+
+    The gateway being alive says nothing about the engine behind a given model, so the
+    health probe checks that model's own upstream. Cached briefly: /model/info lists every
+    model and the status bar polls often. ponytail: duplicate model_name deployments, the
+    last one wins.
+    """
+    if not (api_key and model):
+        return ""
+    now = time.time()
+    cached = _GATEWAY_UPSTREAMS.get(root)
+    if not cached or now - cached[0] > _GATEWAY_UPSTREAM_TTL:
+        mapping: dict[str, str] = {}
+        try:
+            r = requests.get(root + "/model/info", headers={"Authorization": f"Bearer {api_key}"}, timeout=3)
+            if r.status_code == 200:
+                for m in (r.json().get("data") or []):
+                    base = (m.get("litellm_params") or {}).get("api_base")
+                    if m.get("model_name") and base:
+                        mapping[str(m["model_name"])] = str(base)
+        except Exception:
+            pass
+        cached = _GATEWAY_UPSTREAMS[root] = (now, mapping)
+    return cached[1].get(model, "")
+
+
+def _backend_health_detail(url: str, api_key: str = "", model: str = "") -> dict:
+    headers = {"Authorization": f"Bearer {api_key}"} if api_key else {}
+    # Card URLs usually end in /v1 (OpenAI-compatible base); health endpoints live at the
+    # server root, so probe from the root.
+    root = url.rstrip("/").removesuffix("/v1")
+    checks = [("/health", None), ("/v1/health", None), ("/v1/audio/list_voices", None)]
+    upstream = _gateway_upstream(root, api_key, model)
+    if upstream:
+        detail = probe_provider(upstream.rstrip("/").removesuffix("/v1"), checks, timeout=2)
+        # A refused connection is definitive (engine container is down); anything else
+        # (DNS, timeout) may just mean the upstream is not routable from here, so fall back.
+        if detail.get("ok") or detail.get("error") == "Connection refused":
+            return detail
+    # /health/liveliness is the unauthenticated liveness route of a LiteLLM gateway
+    # (its /health and /v1/models need a key): the gateway itself is up.
+    return probe_provider(root, checks + [("/health/liveliness", None)], headers=headers, timeout=2)
 
 
 def _backend_health(url: str) -> bool:
@@ -416,6 +514,28 @@ def sync_voice_clone_reference(audio: Path, transcript: str, language: str, sett
     time.sleep(0.35)
 
 
+def publish_voice_clone_if_missing(audio: Path, language: str, settings: dict) -> bool:
+    """Publish a voice that the engine's voices.json does not know yet (e.g. just unhidden).
+
+    Existing entries are left alone: syncing drops the cached speaker embedding, which
+    would force a needless re-warm. Returns True when a new entry was written.
+    """
+    voices_path = _seed_finder_voices_json_path(settings)
+    if not voices_path.exists():
+        return False
+    try:
+        known = json.loads(voices_path.read_text(encoding="utf-8"))
+    except (OSError, ValueError):
+        return False
+    if not isinstance(known, dict) or audio.stem in known:
+        return False
+    _has_ref, transcript = _read_reference_text(audio)
+    if not transcript:
+        return False
+    sync_voice_clone_reference(audio, transcript, language, settings)
+    return True
+
+
 def _seed_finder_voice_config(voice: str, voices: dict, settings: dict) -> dict:
     cfg = dict(voices.get(voice) or {})
     if cfg.get("ref_audio"):
@@ -573,6 +693,10 @@ def _fetch_backend_voices(settings: dict, backend: str) -> list:
         return _active_library_voice_options(settings)
     tts_url = _validate_http_url(_preview_backend_base_url(settings, backend), allow_private=True).rstrip("/")
     key = (settings.get("voice_design_api_key") if backend == "voice_design" else settings.get("tts_api_key")) or ""
+    from core.provider_health import provider_key_for_url
+    if backend != "voice_design" and tts_url != str(settings.get("tts_url") or "").rstrip("/"):
+        key = ""
+    key = _tts_api_key(settings, tts_url, backend.removeprefix("custom:")) if backend.startswith("custom:") else provider_key_for_url(settings, tts_url, key)
     tts_hdrs = {"Authorization": f"Bearer {key.strip()}"} if key.strip() else {}
     for ep in _TTS_VOICE_ENDPOINTS:
         try:
@@ -643,16 +767,22 @@ async def tts_backends():
     items = []
     for backend in ("voice_clone", "voice_design", "customvoice", "streaming", "fishspeech", "kokoro", "vibevoice", "xtts", "nvidia_magpie", "nvidia_zeroshot", "nvidia_flow"):
         url = _validate_http_url(_preview_backend_base_url(settings, backend), allow_private=True).rstrip("/")
-        voices = _fetch_backend_voices(settings, backend)
+        disabled = (settings.get("engine_disabled") or {}).get(_TTS_BACKEND_CARDS[backend]) is True
-        health_detail = _backend_health_detail(url)
+        setting_key = {"voice_clone": "tts_url", "voice_design": "voice_design_url", "customvoice": "customvoice_url", "streaming": "tts_stream_url", "fishspeech": "fishspeech_url", "nvidia_magpie": "nvidia_tts_url", "nvidia_zeroshot": "nvidia_zeroshot_url", "nvidia_flow": "nvidia_flow_url", "kokoro": "kokoro_url", "vibevoice": "vibevoice_url", "xtts": "xtts_url"}[backend]
+        selected = (settings.get("engine_active") or {}).get(setting_key, "")
+        disabled = disabled or engine_disconnected(settings, url, "tts", selected)
+        voices = [] if disabled else _fetch_backend_voices(settings, backend)
+        health_detail = {"ok": False} if disabled else _backend_health_detail(url)
         health = bool(health_detail.get('ok'))
         available = _backend_available(backend, voices, health)
         items.append({
             "id": backend,
-            "label": _backend_display_name(backend, url),
+            "label": _backend_display_name(backend, url, settings),
+            "title": _backend_card_title(backend),
             "url": url,
             "port": _backend_port_label(url),
             "available": available,
+            "disabled": disabled,
             "configured": _backend_configured(settings, backend, voices),
             "health": {
                 "ok": health,
@@ -663,6 +793,42 @@ async def tts_backends():
             "voice_count": len(voices) if isinstance(voices, list) else 0,
             **_backend_capabilities(backend),
         })
+    for custom in settings.get("custom_engine_cards", []):
+        if custom.get("role") != "tts":
+            continue
+        c_id = custom.get("id") or custom.get("name")
+        backend_id = f"custom:{c_id}"
+        try:
+            url = _validate_http_url(_preview_backend_base_url(settings, backend_id), allow_private=True).rstrip("/")
+        except (ValueError, HTTPException):
+            continue
+        from core.tts_helpers import _tts_selected_model
+        model = _tts_selected_model(settings, str(c_id), "")
+        label = f"{model} (Custom)" if model else f"{custom.get('label') or c_id} (Custom)"
+        disabled = (settings.get("engine_disabled") or {}).get(str(c_id)) is True
+        voices = [] if disabled else _fetch_backend_voices(settings, backend_id)
+        health_detail = {"ok": False} if disabled else _backend_health_detail(url, _tts_api_key(settings, url, str(c_id)), model)
+        health = bool(health_detail.get('ok'))
+        available = _backend_available(backend_id, voices, health)
+        items.append({
+            "id": backend_id,
+            "label": label,
+            "title": custom.get("label") or custom.get("name") or str(c_id),
+            "url": url,
+            "port": _backend_port_label(url),
+            "available": available,
+            "disabled": disabled,
+            "configured": True,
+            "health": {
+                "ok": health,
+                "status": health_detail.get("status"),
+                "endpoint": health_detail.get("endpoint"),
+                "error": health_detail.get("error"),
+            },
+            "voice_count": len(voices) if isinstance(voices, list) else 0,
+            **_backend_capabilities(backend_id),
+        })
+
     return {"backends": items}
 
 
@@ -1214,7 +1380,7 @@ async def openai_speech_proxy(request: Request):
             audio, media_type = await _bounded_tts_call(
                 _tts_request_audio, text, voice, settings, style_instruction,
                 _preview_backend_base_url(settings, "nvidia_magpie"),
-                settings.get("tts_api_key", ""), "nvidia_magpie", "nvidia_magpie",
+                None, "nvidia_magpie", "nvidia_magpie",
             )
         elif backend in {"nvidia_zeroshot", "nvidia_flow"}:
             audio, media_type = await _bounded_tts_call(
@@ -1223,6 +1389,10 @@ async def openai_speech_proxy(request: Request):
                 str(data.get("audio_prompt_transcript") or ""),
                 str(data.get("language") or ""),
             )
+        elif backend.startswith("custom:"):
+            audio, media_type = await _bounded_tts_call(
+                _preview_request_audio, text, voice, settings, style_instruction, backend,
+            )
         elif backend == "fishspeech":
             audio, media_type = await _bounded_tts_call(
                 _fishspeech_request_audio, text, voice, settings,

scripts/sonar_triage.mjs

@@ -73,11 +73,21 @@ export function triage(issues, cfg) {
 
 const countBy = (xs, f) => xs.reduce((m, x) => (m[f(x)] = (m[f(x)] || 0) + 1, m), {});
 
+export const IMPACT_RANK = { BLOCKER: 0, HIGH: 1, MEDIUM: 2, LOW: 3, INFO: 4 };
+export function impactSeverity(issue) {
+  const severities = (issue.impacts || []).map((i) => i.severity);
+  return severities.sort((a, b) => (IMPACT_RANK[a] ?? 5) - (IMPACT_RANK[b] ?? 5))[0];
+}
+export function parseImpactSeverities(value) {
+  const severities = (value || '').split(',');
+  if (!severities.length || severities.some((s) => !Object.hasOwn(IMPACT_RANK, s))) throw new Error('--impact-severities requires BLOCKER,HIGH,MEDIUM,LOW or INFO');
+  return new Set(severities);
+}
 export const MAX_LOCATIONS = 40;
 export const REFACTOR_RULES = new Set(['python:S3776', 'javascript:S3776']);   // cognitive complexity: own lane and agent
 
 // One group per rule; a rule with more than `size` locations is split into numbered parts.
-export function groupIssues(items, size = MAX_LOCATIONS) {
+export function groupIssues(items, size = MAX_LOCATIONS, severityFirst = false) {
   const byRule = new Map();
   for (const i of items) (byRule.get(i.rule) || byRule.set(i.rule, []).get(i.rule)).push(i);
   const out = [];
@@ -99,7 +109,8 @@ export function groupIssues(items, size = MAX_LOCATIONS) {
   const SEV_RANK = { BLOCKER: 0, CRITICAL: 1, MAJOR: 2, MINOR: 3, INFO: 4 };
   const NOISY = new Set(['javascript:S3358', 'python:S1192', 'python:S8415', 'javascript:S5850', 'python:S5850', 'javascript:S2189']);   // triage: low value or probable false positive, file last
   const rank = (g) => [NOISY.has(g.rule) ? 1 : 0, Math.min(...g.issues.map((i) => TYPE_RANK[i.type] ?? 3)), Math.min(...g.issues.map((i) => SEV_RANK[i.severity] ?? 5))];
-  out.sort((a, b) => { const x = rank(a), y = rank(b); return x[0] - y[0] || x[1] - y[1] || x[2] - y[2] || a.rule.localeCompare(b.rule) || a.part - b.part; });
+  out.sort((a, b) => { const x = rank(a), y = rank(b); const impact = (g) => Math.min(...g.issues.map((i) => IMPACT_RANK[impactSeverity(i)] ?? 5));
+    return (severityFirst ? impact(a) - impact(b) : 0) || x[0] - y[0] || x[1] - y[1] || x[2] - y[2] || a.rule.localeCompare(b.rule) || a.part - b.part; });
   return out;
 }
 
@@ -149,10 +160,11 @@ export function postWebhook(url, payload) {
 
 function ghIssues(groups, apply, cap, log, lane = 'fix', noHook = false) {
   const agentLabel = lane === 'refactor' ? 'agent-refactor' : 'agent-fix';
-  if (run('gh', ['auth', 'status']).status !== 0) { log('gh unauthenticated: skipping GitHub issues (run `gh auth login`).'); return []; }
+  if (run('gh', ['auth', 'status']).status !== 0) { if (apply) throw new Error('gh unauthenticated'); log('gh unauthenticated: skipping GitHub issues (run `gh auth login`).'); return []; }
   const list = run('gh', ['issue', 'list', '--state', 'open', '--label', 'sonar', '--limit', '300', '--json', 'number,body']);
   // closed fixed-local issues also suppress (Sonar keys stay open until a re-scan)
   const fixed = run('gh', ['issue', 'list', '--state', 'closed', '--label', 'fixed-local', '--limit', '300', '--json', 'number,body']);
+  if (apply && [list, fixed].some((r) => r.status !== 0)) throw new Error('gh issue list failed');
   const bodies = [list, fixed].flatMap((l) => (l.status === 0 ? JSON.parse(l.stdout).map((x) => x.body) : []));
   const todo = groups.filter((g) => !hasMarker(bodies, g.keys));
   log(`GitHub: ${groups.length - todo.length} duplicate(s) skipped, ${Math.min(todo.length, cap)} of ${todo.length} to create (cap ${cap}).`);
@@ -161,7 +173,7 @@ function ghIssues(groups, apply, cap, log, lane = 'fix', noHook = false) {
   for (const g of todo.slice(0, cap)) {
     if (!apply) { log(`  [dry-run] would create: ${g.title}`); continue; }
     const r = run('gh', ['issue', 'create', '--title', g.title, '--label', 'sonar', '--label', agentLabel, '--body-file', '-'], issueBody(g));
-    if (r.status !== 0) { log(`  gh create failed for ${g.title}`); continue; }
+    if (r.status !== 0) throw new Error(`gh create failed for ${g.title}`);
     const url = r.stdout.trim().split('\n').pop();
     log(`  created ${url}`);
     // Mechanism 1: a multica autopilot webhook (URL is a secret: env MULTICA_WEBHOOK_URL or ~/.config/multica-webhook).
@@ -188,15 +200,15 @@ export async function drain(log, { lane = 'fix', timeoutMin = 60, resends = 1, p
   const sleep = deps.sleep || ((ms) => new Promise((r) => setTimeout(r, ms)));
   const hook = deps.webhookUrl ? deps.webhookUrl() : webhookUrl(process.env, undefined, lane);
   const post = deps.postWebhook || postWebhook;
-  if (!hook) { log('no webhook configured for this lane'); return { done: [], stalled: null }; }
+  if (!hook) throw new Error('no webhook configured for this lane');
   const label = lane === 'refactor' ? 'agent-refactor' : 'agent-fix';
   const listed = sh('gh', ['issue', 'list', '--state', 'open', '--label', 'sonar', '--label', label, '--limit', '100', '--json', 'number,title,body,url']);
-  if (listed.status !== 0) { log('gh issue list failed'); return { done: [], stalled: null }; }
+  if (listed.status !== 0) throw new Error('gh issue list failed');
   const todo = JSON.parse(listed.stdout).sort((x, y) => x.number - y.number);
   const done = [];
-  const view = (n) => JSON.parse(sh('gh', ['issue', 'view', String(n), '--json', 'state,comments,labels']).stdout || '{}');
+  const view = (n) => { const r = sh('gh', ['issue', 'view', String(n), '--json', 'state,comments,labels']); if (r.status !== 0) throw new Error('gh issue view failed'); return JSON.parse(r.stdout); };
   for (const it of todo) {
-    const base = (view(it.number).comments || []).length;
+    view(it.number); // Fail before delivery if GitHub cannot be read.
     let sent = 0, answered = false;
     while (!answered && sent <= resends) {
       const w = post(hook, { source: 'sonar-triage', title: it.title, github_issue_url: it.url, body: it.body });
@@ -206,7 +218,7 @@ export async function drain(log, { lane = 'fix', timeoutMin = 60, resends = 1, p
       while (Date.now() < deadline) {
         await sleep(pollSec * 1000);
         const v = view(it.number);
-        if (v.state === 'CLOSED' || (v.comments || []).length > base || (v.labels || []).some((l) => l.name === 'fixed-local')) { answered = true; break; }
+        if (v.state === 'CLOSED' || (v.labels || []).some((l) => l.name === 'fixed-local')) { answered = true; break; }
       }
       if (!answered) log(`#${it.number}: no answer within ${timeoutMin} min`);
     }
@@ -247,6 +259,7 @@ export async function main(argv, deps = {}) {
     else if (f === '--no-webhook') a.noHook = true;
     else if (f === '--timeout') a.timeout = +argv[++i];
     else if (f === '--resends') a.resends = +argv[++i];
+    else if (f === '--impact-severities') a.impactSeverities = parseImpactSeverities(argv[++i]);
     else if (f === '--exclude-rules') a.exclude = (argv[++i] || '').split(',').filter(Boolean);
     else throw new Error(`unknown flag ${f}`);
   }
@@ -254,13 +267,16 @@ export async function main(argv, deps = {}) {
     if (!a.apply) { log('--drain needs --apply (it fires webhooks).'); return {}; }
     return drain(log, { lane: a.lane || 'fix', timeoutMin: a.timeout || 60, resends: a.resends ?? 1, deps: deps.drainDeps });
   }
+  if (a.impactSeverities && (a.exclude?.length || a.lane === 'refactor' || a.fixSmall)) throw new Error('--impact-severities cannot be combined with exclusions, refactor lane or --fix-small');
   const cfg = deps.cfg || loadConfig();
   if (a.list) { log('Allowlisted (small) rules:'); for (const r of cfg.allow) log(`  ${r}`); log('Skip-list:'); for (const s of cfg.skip) log(`  ${s.rule || '*'} ${s.file || ''} ${s.message || ''} -- ${s.reason}`); return; }
   const tokFile = join(homedir(), '.config', 'sonar-token');
   const token = deps.token || process.env.SONAR_TOKEN || (existsSync(tokFile) ? readFileSync(tokFile, 'utf8').trim() : '');
   if (!token) throw new Error('no token: put it in ~/.config/sonar-token (or SONAR_TOKEN)');
-  const issues = await fetchIssues({ fetchFn: deps.fetchFn, token, newCode: a.newCode, max: a.max });
+  const fetched = await fetchIssues({ fetchFn: deps.fetchFn, token, newCode: a.newCode, max: a.max });
+  const issues = a.impactSeverities ? fetched.filter((i) => a.impactSeverities.has(impactSeverity(i))) : fetched;
   const res = triage(issues, cfg);
+  if (a.impactSeverities) log(`Selected impact findings: ${issues.length}; scan freshness has not been verified.`);
   log(`Mode: ${a.apply ? 'APPLY' : 'dry-run'} | scope: ${a.newCode ? 'new code' : 'all'} | issues: ${issues.length}`);
   for (const c of ['small', 'medium', 'large', 'skipped']) {
     log(`\n${c.toUpperCase()} (${res[c].length})`);
@@ -270,13 +286,13 @@ export async function main(argv, deps = {}) {
   if (deps.noActions) return res;
   const lane = a.lane || 'fix';
   // refactor lane: cognitive-complexity findings only, one GitHub issue per function (reviewed and reverted per function), cap 3.
-  const all = [...res.medium, ...res.large];
+  const all = a.impactSeverities ? Object.values(res).flat() : [...res.medium, ...res.large];
   const groups = (lane === 'refactor'
     ? groupIssues(all.filter((i) => REFACTOR_RULES.has(i.rule)), 1)
-    : groupIssues(all)).filter((g) => !(a.exclude || []).includes(g.rule));
+    : groupIssues(all, MAX_LOCATIONS, !!a.impactSeverities)).filter((g) => !(a.exclude || []).includes(g.rule));
   const cap = lane === 'refactor' && !a.capSet ? 3 : a.cap;
   log(`\n${lane === 'refactor' ? 'Refactor' : 'Medium+'} groups (${lane === 'refactor' ? 'one per function' : 'per rule, max 40 locations'}): ${groups.length}`);
-  ghIssues(groups, a.apply, cap, log, lane, !!a.noHook);
+  (deps.ghIssues || ghIssues)(groups, a.apply, cap, log, lane, !!a.noHook);
   if (a.fixSmall) { if (a.apply) fixSmall(res.small, log); else log('fix-small needs --apply (dry-run).'); }
   return res;
 }

static/js/ai-backends.js

@@ -70,20 +70,12 @@
       card?.classList.add('llm-local-card-online');
     }
     btn.addEventListener('click', async () => {
-      if (btn.classList.contains('ok')) {
-        btn.innerHTML = '<span class="mdi mdi-lan-connect"></span> Connect';
-        btn.className = 'llm-cloud-connect-btn llm-local-ping';
-        card?.classList.remove('llm-local-card-online');
-        localStorage.removeItem('llm-cloud-con-' + key);
-        return;
-      }
       const apiKeyInp = card?.querySelector(`.llm-input[data-llm-key="${CSS.escape(key)}"]`);
       const apiKey = apiKeyInp?.value.trim() || '';
-      if (!apiKey) { toast('Enter an API key first', 'error'); return; }
       btn.disabled = true;
       btn.innerHTML = '<span class="mdi mdi-loading mdi-spin"></span> Connecting…';
       try {
-        const d = await probeUrl(endpoint, probeType, apiKey);
+        const d = await probeUrl(endpoint, probeType, apiKey, key);
         if (d.ok) {
           btn.innerHTML = '<span class="mdi mdi-check-network"></span> Connected';
           btn.className = 'llm-local-ping llm-cloud-connect-btn ok';
@@ -114,7 +106,6 @@
       const card = btn.closest('.llm-card');
       const apiKeyInp = card?.querySelector(`.llm-input[data-llm-key="${CSS.escape(key)}"]`);
       const apiKey = apiKeyInp?.value.trim() || '';
-      if (!apiKey) { toast('Enter an API key first', 'error'); return; }
       if (btn.dataset.llmIncompatible && typeof confirmDialog === 'function') {
         const proceed = await confirmDialog(
           "Anthropic doesn't speak the OpenAI-compatible /chat/completions format every LLM call in this app uses — setting it active will very likely break generation until Anthropic support is added. Set it anyway?",
@@ -122,13 +113,20 @@
         );
         if (!proceed) return;
       }
-      const patch = { llm_url: url, llm_api_key: apiKey };
+      try {
-      if (model) patch.llm_model = model;
+        const headers = { 'Content-Type': 'application/json' };
-      await applyAndSaveSettings(patch);
+        if (apiKey) headers['X-Provider-API-Key'] = apiKey;
-      document.querySelectorAll('.llm-cloud-use-btn.active').forEach(b => b.classList.remove('active'));
+        const response = await fetch(`/api/local-containers/${encodeURIComponent(key)}/use-as-llm`, {
-      btn.classList.add('active');
+          method: 'POST', headers, body: JSON.stringify({ url, model: card?.querySelector('.cloud-model-select')?.value || model }),
-      window.syncActiveLlmPanel?.(url, apiKey);
+        });
-      toast(`✓ Active LLM set to ${card?.querySelector('.llm-card-name')?.textContent || key}`, 'success');
+        const result = await response.json();
+        if (!response.ok || !result.ok) { toast(result.error || 'Could not activate LLM', 'error'); return; }
+        Object.assign(_appSettings || (_appSettings = {}), await fetch('/api/settings').then(r => r.json()));
+        document.querySelectorAll('.llm-cloud-use-btn.active').forEach(b => b.classList.remove('active'));
+        btn.classList.add('active');
+        window.syncActiveLlmPanel?.(url, apiKey);
+        toast(`✓ Active LLM set to ${card?.querySelector('.llm-card-name')?.textContent || key}`, 'success');
+      } catch (e) { toast('Could not activate LLM: ' + e.message, 'error'); }
     });
   });
 
@@ -192,7 +190,7 @@
         const type = cardType(card);
         const apiKey = typeof window.engineApiKeyValue === 'function'
           ? window.engineApiKeyValue(key) : '';
-        const d = await probeUrl(rawUrl, type, apiKey);
+        const d = await probeUrl(rawUrl, type, apiKey, key);
         applyCardState(card, key, d.ok, !d.ok);
         if (d.ok) {
           toast(`✓ ${type.toUpperCase() || 'Service'} reachable — ${d.endpoint}`, 'success');
@@ -230,6 +228,109 @@
   window.applyAndSaveSettings = applyAndSaveSettings;
   window.probeUrl = probeUrl;
 
+  async function fetchEngineModels(url, type, apiKey = '', name = '') {
+    const params = new URLSearchParams({ url, type, name });
+    return fetch('/api/engine-models?' + params, {
+      headers: apiKey ? { 'X-Provider-API-Key': apiKey } : {},
+    }).then(r => r.json());
+  }
+
+  document.querySelectorAll('.llm-card').forEach(card => {
+    const keyInp = card.querySelector('.llm-input[data-llm-key]');
+    if (!keyInp) return;
+    const key = keyInp.dataset.llmKey;
+    const useBtn = card.querySelector('.llm-cloud-use-btn');
+    const groqStt = key === 'groq_stt';
+    if (!useBtn && !groqStt) {
+      if (['hf_stt', 'assemblyai', 'fish_audio'].includes(key)) {
+        const note = document.createElement('div');
+        note.className = 'llm-info-note';
+        note.textContent = 'Provider-native transcription or speech API; connection testing and model discovery are not integrated.';
+        card.appendChild(note);
+      }
+      return;
+    }
+    const url = useBtn?.dataset.llmEndpoint || 'https://api.groq.com/openai/v1';
+    const type = groqStt ? 'stt' : useBtn.dataset.llmProbeType || (useBtn.dataset.llmIncompatible ? 'anthropic' : 'llm');
+    const controls = document.createElement('div');
+    controls.className = 'llm-field-row';
+    controls.innerHTML = '<select class="llm-input cloud-model-select" aria-label="Provider model"><option value="">— load models —</option></select><button type="button" class="llm-use-btn">Load models</button>';
+    const select = controls.querySelector('select');
+    const loadBtn = controls.querySelector('button');
+    const wanted = _appSettings?.engine_models?.[key] || useBtn?.dataset.llmModel || '';
+    if (wanted) select.innerHTML = `<option value="${escHtml(wanted)}">${escHtml(wanted)}</option>`;
+    loadBtn.addEventListener('click', async () => {
+      loadBtn.disabled = true;
+      try {
+        const d = await fetchEngineModels(url, type, keyInp.value.trim(), key);
+        if (!d.ok) { toast(d.error || 'Could not load models', 'error'); return; }
+        const old = select.value;
+        const models = d.models || [];
+        select.innerHTML = '<option value="">— select a model —</option>' + models.map(m => `<option value="${escHtml(m)}">${escHtml(m)}</option>`).join('');
+        if (models.includes(old)) select.value = old;
+      } catch (e) { toast('Could not load models: ' + e.message, 'error'); }
+      finally { loadBtn.disabled = false; }
+    });
+    select.addEventListener('change', () => applyAndSaveSettings({ engine_models: { ...(_appSettings?.engine_models || {}), [key]: select.value } }));
+    if (groqStt) {
+      const testBtn = document.createElement('button');
+      testBtn.type = 'button'; testBtn.className = 'llm-use-btn'; testBtn.textContent = 'Test connection';
+      testBtn.addEventListener('click', async () => {
+        testBtn.disabled = true;
+        try {
+          const d = await probeUrl(url, 'stt', keyInp.value.trim(), key);
+          toast(d.ok ? 'Service reachable' : d.error || 'Connection failed', d.ok ? 'success' : 'error');
+        } catch (e) { toast('Connection failed: ' + e.message, 'error'); }
+        finally { testBtn.disabled = false; }
+      });
+      const activateBtn = document.createElement('button');
+      activateBtn.type = 'button'; activateBtn.className = 'llm-use-btn'; activateBtn.textContent = 'Use as STT';
+      activateBtn.addEventListener('click', async () => {
+        const headers = { 'Content-Type': 'application/json' };
+        if (keyInp.value.trim()) headers['X-Provider-API-Key'] = keyInp.value.trim();
+        try {
+          const d = await fetch(`/api/local-containers/${key}/use-as-engine`, { method: 'POST', headers,
+            body: JSON.stringify({ url, role: 'stt', setting_key: 'whisper_url', model: select.value }) }).then(r => r.json());
+          if (d.ok) {
+            Object.assign(_appSettings || (_appSettings = {}), await fetch('/api/settings').then(r => r.json()));
+            $('s-whisper-url').value = url;
+          }
+          toast(d.ok ? 'Active STT updated' : d.error || 'Activation failed', d.ok ? 'success' : 'error');
+        } catch (e) { toast('Activation failed: ' + e.message, 'error'); }
+      });
+      controls.append(testBtn, activateBtn);
+    }
+    card.appendChild(controls);
+  });
+
+  ['s-stt-test-connection', 's-stt-load-models'].forEach(id => {
+    $(id)?.addEventListener('click', async () => {
+      const url = $('s-whisper-url').value.trim();
+      const apiKey = $('s-stt-api-key').value.trim();
+      const status = $('s-stt-connection-status');
+      const btn = $(id);
+      if (!url) { status.textContent = 'Enter an endpoint first'; return; }
+      btn.disabled = true;
+      status.textContent = 'Checking…';
+      try {
+        const d = id === 's-stt-load-models'
+          ? await fetchEngineModels(url, 'stt', apiKey)
+          : await probeUrl(url, 'stt', apiKey);
+        if (!d.ok) { status.textContent = d.error || 'Connection failed'; return; }
+        status.textContent = 'Service reachable';
+        if (id === 's-stt-load-models') {
+          const select = $('s-stt-model');
+          const wanted = select.value;
+          const models = d.models || [];
+          select.innerHTML = '<option value="">— select a model —</option>' + models.map(m => `<option value="${escHtml(m)}">${escHtml(m)}</option>`).join('');
+          if (models.includes(wanted)) select.value = wanted;
+          status.textContent = models.length ? `${models.length} models loaded` : 'No models advertised';
+        }
+      } catch (e) { status.textContent = 'Connection failed: ' + e.message; }
+      finally { btn.disabled = false; }
+    });
+  });
+
   $('llm-use-faster-whisper-stt')?.addEventListener('click', () => {
     const url = document.querySelector('[data-llm-local-key="faster-whisper"]')?.value.trim()
       || 'http://host.docker.internal:8000';
@@ -315,6 +416,14 @@
     };
   }
 
+  function activeEngineDisconnected(settingKey, url, modelSel, setStatus) {
+    const identity = _appSettings?.engine_active?.[settingKey];
+    if (url !== _appSettings?.[settingKey] || _appSettings?.engine_disabled?.[identity] !== true) return false;
+    modelSel.innerHTML = '<option value="">Disconnected</option>';
+    setStatus('Disconnected', 'err');
+    return true;
+  }
+
   // ── Active Language Model (global picker) ───────────────────────────────
   (function initActiveLlmPanel() {
     upgradeModelSelect('llm-active-model');
@@ -329,12 +438,13 @@
     async function fetchModels(url, apiKey) {
       url = (url || urlInp.value.trim());
       if (!url) { modelSel.innerHTML = '<option value="">— set an endpoint —</option>'; setStatus(''); return; }
+      if (activeEngineDisconnected('llm_url', url, modelSel, setStatus)) return;
       const want = (_appSettings && _appSettings.llm_model) || '';
       setStatus('Loading models…');
       if (refreshBtn) refreshBtn.disabled = true;
       try {
-        const fetchUrl = '/api/conversation/llm-models?url=' + encodeURIComponent(url);
+        const d = await fetchEngineModels(url, 'llm', apiKey);
-        const d = await fetch(fetchUrl).then(r => r.json());
+        if (activeEngineDisconnected('llm_url', url, modelSel, setStatus)) return;
         if (d.error) {
            let msg = 'Failed';
            try { 
@@ -468,11 +578,13 @@
     async function fetchModels(url) {
       url = (url || urlInp.value.trim());
       if (!url) { modelSel.innerHTML = '<option value="">— set an endpoint —</option>'; setStatus(''); return; }
+      if (activeEngineDisconnected('vision_ocr_url', url, modelSel, setStatus)) return;
       const want = (_appSettings && _appSettings.vision_ocr_model) || '';
       setStatus('Loading models…');
       if (refreshBtn) refreshBtn.disabled = true;
       try {
-        const d = await fetch('/api/conversation/llm-models?url=' + encodeURIComponent(url)).then(r => r.json());
+        const d = await fetchEngineModels(url, 'ocr');
+        if (activeEngineDisconnected('vision_ocr_url', url, modelSel, setStatus)) return;
         if (d.error) {
           let msg = d.error;
           try { const j = JSON.parse(d.error); msg = j.error?.message || j.detail || d.error; } catch (_) {}

static/js/conversation.js

@@ -335,17 +335,17 @@ $('s-import-voices-file')?.addEventListener('change', async function () {
   // ── Populate STT backends ────────────────────────────────────────────────
   async function loadConvSttBackends() {
     if (!sttSel) return;
-    const prev = sttSel.value;
+    const prev = _appSettings?.stt_preferred_backend || sttSel.value;
     try {
       const d = await fetch('/api/stt-backends').then(r => r.json());
-      const all = d.backends || [];
+      const all = sttBackendChoices(d.backends || [], prev);
       if (!all.length) {
         sttSel.innerHTML = '<option value="configured">Default (configured)</option>';
         return;
       }
       sttSel.innerHTML = all.map(b => {
         const icon = b.available ? '✓' : '✗';
-        const label = `${icon} ${escHtml(b.label)}`;
+        const label = `${icon} ${escHtml(b.title || b.label || b.id)}`;
         const disabled = !b.available;
         return `<option value="${escHtml(b.id)}"${disabled ? ' disabled' : ''}>${label}</option>`;
       }).join('');

static/js/engines.js

@@ -253,6 +253,7 @@ function loadCustomEngineCards() {
   return merged;
 }
 function saveCustomEngineCards(cards) {
+  cards = cards.map(({ apiKey, api_key, ...card }) => card);
   localStorage.setItem('engines-custom-cards', JSON.stringify(cards));
   if (_appSettings) _appSettings.custom_engine_cards = cards;
   _patchSettings({ custom_engine_cards: cards });
@@ -288,26 +289,107 @@ const DC_ROLE_USE = {
   ocr: { settingKey: 'vision_ocr_url',       label: 'Use as OCR' },
 };
 
+async function activateEngine(id, role, settingKey, url, model, apiKey = '') {
+  const headers = { 'Content-Type': 'application/json' };
+  if (apiKey) headers['X-Provider-API-Key'] = apiKey;
+  const body = role === 'llm' ? { url, model } : { url, model, role, setting_key: settingKey, activate: true };
+  const response = await fetch(`/api/local-containers/${encodeURIComponent(id)}/use-as-${role === 'llm' ? 'llm' : 'engine'}`, {
+    method: 'POST', headers, body: JSON.stringify(body),
+  });
+  const result = await response.json();
+  if (!response.ok || !result.ok) throw new Error(result.error || 'Activation failed');
+  _appSettings[settingKey] = url;
+  try { Object.assign(_appSettings, await fetch('/api/settings').then(r => r.json())); } catch (_) { /* Activation is already saved. */ }
+  try {
+    if (role === 'tts') {
+      const custom = (_appSettings.custom_engine_cards || []).some(card => String(card.id || card.name) === id);
+      await window.refreshTtsBackendAvailability?.(custom ? 'custom:' + id : '');
+    }
+    if (role === 'stt') {
+      await window.refreshSttBackends?.(_appSettings.stt_preferred_backend || '');
+      window.syncPreferredSttBackend?.(_appSettings);
+    }
+  } catch (_) { /* Optional picker refresh does not undo saved activation. */ }
+  if (role === 'llm') window.syncActiveLlmPanel?.(url);
+  if (role === 'ocr') window.syncActiveOcrPanel?.(url);
+  try { await window.refreshStatusBarEngines?.({ force: true }); } catch (_) { /* status bar catches up on its next poll */ }
+}
+
+async function setEngineConnection(id, connected, role) {
+  const response = await fetch(`/api/local-containers/${encodeURIComponent(id)}/connection`, {
+    method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ connected }),
+  });
+  const result = await response.json();
+  if (!response.ok || !result.ok) throw new Error(result.error || 'Connection change failed');
+  const disabled = { ...(_appSettings?.engine_disabled || {}) };
+  if (connected) delete disabled[id]; else disabled[id] = true;
+  _appSettings.engine_disabled = disabled;
+  try {
+    Object.assign(_appSettings, await fetch('/api/settings').then(r => r.json()));
+    await window.refreshTtsBackendAvailability?.();
+    if (role === 'stt') {
+      await window.refreshSttBackends?.();
+      window.syncPreferredSttBackend?.(_appSettings);
+    }
+    if (role === 'llm') window.syncActiveLlmPanel?.(_appSettings.llm_url || '');
+    if (role === 'ocr') window.syncActiveOcrPanel?.(_appSettings.vision_ocr_url || '');
+    await window.refreshStatusBarEngines?.({ force: true });
+  } catch (_) { /* The connection change is already saved. */ }
+}
+
+let _inlineEngineCard = null;
+
+function closeEngineEditor() {
+  const dlg = $('add-engine-dlg');
+  if (_inlineEngineCard) {
+    dlg.appendChild(_inlineEngineCard.querySelector('.add-engine-form'));
+    _inlineEngineCard.classList.remove('dc-editing');
+    const editButton = _inlineEngineCard.querySelector('.dc-edit-btn');
+    _inlineEngineCard = null;
+    editButton?.focus();
+  }
+  dlg.dataset.requestId = String(Number(dlg.dataset.requestId || 0) + 1);
+  delete dlg.dataset.editId;
+  $('aed-api-key').value = '';
+  dlg.close();
+}
+
 window.openAddEngineDialog = function(role) {
+  if ($('aed-save')?.disabled) return;
   const dlg = document.getElementById('add-engine-dlg');
   if (!dlg) return;
-  delete dlg.dataset.editId;
+  closeEngineEditor();
+  dlg.dataset.cardId = 'custom-' + (crypto.randomUUID?.() || Date.now());
+  dlg.dataset.requestId = String(Number(dlg.dataset.requestId || 0) + 1);
+  delete dlg.dataset.probeName;
   const titleEl = dlg.querySelector('.add-engine-title');
   if (titleEl) titleEl.textContent = ' Add Custom Engine';
   const saveBtn = document.getElementById('aed-save');
   if (saveBtn) saveBtn.textContent = 'Add Card';
   const roleEl = document.getElementById('aed-role');
   if (roleEl && role) roleEl.value = role;
-  ['aed-name', 'aed-url', 'aed-container', 'aed-desc'].forEach(id => {
+  ['aed-name', 'aed-url', 'aed-container', 'aed-desc', 'aed-api-key', 'aed-model'].forEach(id => {
     const el = document.getElementById(id); if (el) el.value = '';
   });
+  $('aed-api-key').placeholder = 'API key (optional)';
+  $('aed-models').innerHTML = '';
+  $('aed-status').textContent = '';
+  $('aed-params').value = '{}';
+  $('aed-active').checked = false;
   dlg.showModal();
 };
 
 function editCustomEngineCard(card) {
+  if ($('aed-save')?.disabled) return;
   const dlg = document.getElementById('add-engine-dlg');
   if (!dlg) return;
+  closeEngineEditor();
+  const cardElement = Array.from(document.querySelectorAll('.dc-edit-btn')).find(button => button.dataset.dcCustomId === String(card.id || card.name))?.closest('.llm-local-card');
+  if (!cardElement) return;
   dlg.dataset.editId = String(card.id || card.name);
+  dlg.dataset.cardId = dlg.dataset.editId;
+  dlg.dataset.requestId = String(Number(dlg.dataset.requestId || 0) + 1);
+  dlg.dataset.probeName = _appSettings?.engine_api_keys?.[dlg.dataset.cardId] ? dlg.dataset.cardId : card.name;
   const titleEl = dlg.querySelector('.add-engine-title');
   if (titleEl) titleEl.textContent = ' Edit Engine';
   const saveBtn = document.getElementById('aed-save');
@@ -316,18 +398,76 @@ function editCustomEngineCard(card) {
   if (roleEl) roleEl.value = card.role || 'tts';
   const setVal = (id, val) => { const el = document.getElementById(id); if (el) el.value = val || ''; };
   setVal('aed-name',      card.label || card.name);
-  setVal('aed-url',       card.url);
+  setVal('aed-url',       cardElement.querySelector('.dc-url-inp')?.value || card.url);
-  setVal('aed-container', card.containerName);
+  setVal('aed-container', cardElement.querySelector('.dc-container-name-inp')?.value ?? card.containerName);
   setVal('aed-desc',      card.description);
-  dlg.showModal();
+  setVal('aed-api-key', '');
+  setVal('aed-model', _appSettings?.engine_models?.[dlg.dataset.cardId] || _appSettings?.engine_models?.[card.name]);
+  $('aed-api-key').placeholder = _appSettings?.engine_api_keys?.[dlg.dataset.cardId] || _appSettings?.engine_api_keys?.[card.name]
+    ? 'Configured — enter a new key to replace' : 'API key (optional)';
+  $('aed-models').innerHTML = '';
+  $('aed-status').textContent = '';
+  $('aed-params').value = JSON.stringify(card.params || {}, null, 2);
+  $('aed-active').checked = card.role === 'stt'
+    ? _appSettings?.stt_preferred_backend === 'custom:' + dlg.dataset.cardId
+    : _appSettings?.[DC_ROLE_USE[card.role]?.settingKey] === card.url && (card.role !== 'ocr' || !!_appSettings?.vision_ocr_enabled);
+  _inlineEngineCard = cardElement;
+  cardElement.classList.add('dc-editing');
+  cardElement.appendChild(dlg.querySelector('.add-engine-form'));
+  $('aed-name').focus();
 }
 
 (function initCustomEngineDialog() {
   const dlg = document.getElementById('add-engine-dlg');
   if (!dlg) return;
-  document.getElementById('aed-cancel')?.addEventListener('click', () => { delete dlg.dataset.editId; dlg.close(); });
+  function readParams() {
-  dlg.addEventListener('click', e => { if (e.target === dlg) { delete dlg.dataset.editId; dlg.close(); } });
+    let params;
-  document.getElementById('aed-save')?.addEventListener('click', () => {
+    try { params = JSON.parse($('aed-params').value.trim() || '{}'); }
+    catch (_) { throw new Error('Additional parameters must be valid JSON'); }
+    if (!params || Array.isArray(params) || typeof params !== 'object') throw new Error('Additional parameters must be a JSON object');
+    const reserved = new Set(['model', 'messages', 'input', 'text', 'voice', 'stream', 'response_format', 'file', 'files', 'audio', 'audio_prompt', 'audio_prompt_transcript', 'references', 'url', 'base_url', 'endpoint', 'headers', 'method']);
+    if (Object.keys(params).some(key => reserved.has(key))) throw new Error('Additional parameters cannot replace connection, model, or request input fields');
+    function hasCredentials(value) {
+      return value && typeof value === 'object' && Object.entries(value).some(([key, child]) =>
+        /(apikeys?|authorization|password|passwd|secret|secretkey|accesskey|token|privatekey|credentials?|headers)$/.test(key.toLowerCase().replace(/[^a-z]/g, '')) || hasCredentials(child));
+    }
+    if (hasCredentials(params)) throw new Error('Keep credentials in the API key field, outside additional parameters');
+    return params;
+  }
+  document.getElementById('aed-cancel')?.addEventListener('click', () => { if (!$('aed-save').disabled) closeEngineEditor(); });
+  dlg.addEventListener('click', e => { if (e.target === dlg && !$('aed-save').disabled) closeEngineEditor(); });
+  dlg.addEventListener('cancel', e => { if ($('aed-save').disabled) e.preventDefault(); });
+  ['aed-url', 'aed-role', 'aed-api-key'].forEach(id => $(id)?.addEventListener('input', () => {
+    dlg.dataset.requestId = String(Number(dlg.dataset.requestId || 0) + 1);
+    $('aed-models').innerHTML = '';
+    $('aed-status').textContent = '';
+  }));
+  ['aed-test', 'aed-load-models'].forEach(id => $(id)?.addEventListener('click', async () => {
+    const requestId = dlg.dataset.requestId = String(Number(dlg.dataset.requestId || 0) + 1);
+    const url = $('aed-url').value.trim();
+    if (!url) { $('aed-status').textContent = 'URL is required'; return; }
+    try { readParams(); } catch (e) { $('aed-status').textContent = e.message; return; }
+    const apiKey = $('aed-api-key').value.trim();
+    $(id).disabled = true;
+    $('aed-status').textContent = 'Checking…';
+    try {
+      const route = id === 'aed-test' ? '/api/probe-url?' : '/api/engine-models?';
+      const response = await fetch(route + new URLSearchParams({ url, type: $('aed-role').value, name: dlg.dataset.probeName || dlg.dataset.cardId }), {
+        headers: apiKey ? { 'X-Provider-API-Key': apiKey } : {},
+      });
+      const result = await response.json();
+      if (!(dlg.open || _inlineEngineCard) || dlg.dataset.requestId !== requestId) return;
+      if (!response.ok || !result.ok) { $('aed-status').textContent = result.error || 'Request failed'; return; }
+      $('aed-status').textContent = 'Service reachable';
+      if (id === 'aed-load-models') {
+        const models = result.models || [];
+        $('aed-models').innerHTML = models.map(model => `<option value="${escHtml(model)}"></option>`).join('');
+        $('aed-status').textContent = models.length ? `${models.length} models loaded — select or enter an ID` : 'No models advertised — enter an ID manually';
+      }
+    } catch (e) { if ((dlg.open || _inlineEngineCard) && dlg.dataset.requestId === requestId) $('aed-status').textContent = 'Request failed: ' + e.message; }
+    finally { $(id).disabled = false; }
+  }));
+  document.getElementById('aed-save')?.addEventListener('click', async () => {
     const name          = document.getElementById('aed-name')?.value.trim();
     const role          = document.getElementById('aed-role')?.value;
     const url           = document.getElementById('aed-url')?.value.trim();
@@ -335,27 +475,47 @@ function editCustomEngineCard(card) {
     const desc          = document.getElementById('aed-desc')?.value.trim() || '';
     if (!name) { toast('Name is required', 'error'); return; }
     if (!url)  { toast('URL is required', 'error'); return; }
+    let params;
+    try { params = readParams(); } catch (e) { $('aed-status').textContent = e.message; return; }
     const editId = dlg.dataset.editId;
-    if (editId) {
+    const id = dlg.dataset.cardId || editId || 'custom-' + Date.now();
-      const cards = loadCustomEngineCards().map(c =>
+    const cards = editId ? loadCustomEngineCards().map(c =>
         String(c.id || c.name) === editId
-          ? { ...c, name, label: name, role, url, containerName, description: desc }
+          ? { ...c, id, name, label: name, role, url, containerName, description: desc, params }
           : c
-      );
+      ) : [...loadCustomEngineCards(), { id, name, label: name, role, url, containerName, description: desc, params, isCustom: true }];
-      saveCustomEngineCards(cards);
+    const apiKey = $('aed-api-key').value.trim();
-      delete dlg.dataset.editId;
+    const model = $('aed-model').value.trim();
-      dlg.close();
+    const useActive = $('aed-active').checked;
+    const metadata = cards.map(({ apiKey, api_key, ...card }) => card);
+    const patch = {
+      custom_engine_cards: metadata,
+      engine_models: { ...(_appSettings?.engine_models || {}), [id]: model },
+      engine_local_urls: { ...(_appSettings?.engine_local_urls || {}), ['dc-' + id]: url },
+    };
+    if (apiKey) patch.engine_api_keys = { [id]: apiKey };
+    $('aed-save').disabled = true;
+    $('aed-cancel').disabled = true;
+    try {
+      const response = await fetch('/api/settings', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(patch) });
+      if (!response.ok) throw new Error('Could not save engine');
+      localStorage.setItem('engines-custom-cards', JSON.stringify(metadata));
+      localStorage.setItem('dc-url-' + id, url);
+      Object.assign(_appSettings || (_appSettings = {}), { custom_engine_cards: metadata, engine_models: patch.engine_models, engine_local_urls: patch.engine_local_urls });
+      if (apiKey) _appSettings.engine_api_keys = { ...(_appSettings.engine_api_keys || {}), [id]: '__configured__' };
+      try { Object.assign(_appSettings, await fetch('/api/settings').then(r => r.json())); } catch (_) { /* The save already succeeded. */ }
+      $('aed-api-key').value = '';
+      closeEngineEditor();
       loadLocalContainers();
-      toast(`"${name}" updated`, 'success');
+      toast(`"${name}" ${editId ? 'updated' : 'added'}`, 'success');
-    } else {
+      if (useActive) {
-      const card = { id: 'custom-' + Date.now(), name, label: name, role, url, containerName, description: desc, isCustom: true };
+        try {
-      const cards = loadCustomEngineCards();
+          await activateEngine(id, role, DC_ROLE_USE[role].settingKey, url, model, apiKey);
-      cards.push(card);
+          toast(`"${name}" is active`, 'success');
-      saveCustomEngineCards(cards);
+        } catch (_) { toast(`"${name}" saved; activation failed. Use the card's activation button to retry.`, 'error'); }
-      dlg.close();
+      }
-      loadLocalContainers();
+    } catch (e) { toast(e.message, 'error'); }
-      toast(`"${name}" added`, 'success');
+    finally { $('aed-save').disabled = false; $('aed-cancel').disabled = false; }
-    }
   });
 })();
 
@@ -400,6 +560,7 @@ async function fetchLocalContainersWithRetry() {
 }
 
 async function loadLocalContainers() {
+  if (_inlineEngineCard) return;
   const seq = ++_localContainersLoadSeq;
   const gridTts = $('dc-grid-tts');
   const gridStt = $('dc-grid-stt');
@@ -438,6 +599,8 @@ async function loadLocalContainers() {
 window.loadLocalContainers = loadLocalContainers;
 
 function renderLocalContainers(containers) {
+  if (_inlineEngineCard) return;
+  containers = containers.map(c => c.isCustom ? { ...c, legacyName: c.name, name: c.id || c.name, label: c.label || c.name } : c);
   const gridTts = $('dc-grid-tts');
   const gridStt = $('dc-grid-stt');
   const gridLlm = $('dc-grid-llm');
@@ -524,6 +687,7 @@ function renderLocalContainers(containers) {
     <input class="llm-local-url-inp dc-url-inp" type="text" spellcheck="false"
       placeholder="${escHtml(defaultUrl)}"
       data-dc-url-key="${n}"
+      data-dc-legacy-name="${escHtml(c.legacyName || c.name)}"
       data-dc-url-default="${escHtml(defaultUrl)}">
   </div>`;
 
@@ -541,27 +705,36 @@ function renderLocalContainers(containers) {
       data-cn-key="${n}">
   </div>`;
 
-    // API key row — only for LLMs
+    // Authentication is optional for every supported engine role.
     const savedApiKey = isCustom
-      ? ((typeof window.engineApiKeyValue === 'function' ? window.engineApiKeyValue(c.name, 'dc-apikey-' + c.name) : '') || c.apiKey || '')
+      ? ((typeof window.engineApiKeyValue === 'function' ? window.engineApiKeyValue(c.name, 'dc-apikey-' + c.name) : '') || '')
       : ((typeof window.engineApiKeyValue === 'function' ? window.engineApiKeyValue(c.name) : '') || '');
-    const apiKeyRowHtml = (c.role === 'llm' || c.name === 'litellm') ? `
+    const apiKeyRowHtml = `
   <div class="llm-local-url">
     <span class="llm-local-url-label" title="API Key"><span class="mdi mdi-key-outline"></span></span>
     <input class="llm-local-url-inp dc-apikey-inp" type="password"
       placeholder="API key (optional)" spellcheck="false"
       value="${escHtml(savedApiKey)}"
       data-dc-apikey-key="${n}">
-  </div>` : '';
+  </div>`;
+    const savedModel = _appSettings?.engine_models?.[c.name] || _appSettings?.engine_models?.[c.legacyName] || '';
+    const modelRowHtml = `
+  <div class="llm-local-url">
+    <span class="llm-local-url-label">Model</span>
+    <input class="llm-local-url-inp dc-model-inp" list="dc-models-${n}" aria-label="${escHtml(c.label || c.name)} model" value="${escHtml(savedModel)}" placeholder="Model ID">
+    <datalist id="dc-models-${n}"></datalist>
+    <button class="llm-use-btn dc-models-btn" type="button">Load models</button>
+  </div>`;
 
     // Use-as button — right-aligned, clearly labelled as the activation action
     const useEntry = DC_USE_MAP[c.name] || DC_ROLE_USE[c.role];
+    const useLabel = c.role === 'tts' ? 'Use as TTS' : useEntry?.label;
     const useBtn = useEntry
       ? `<button class="llm-use-btn dc-use-btn"
            data-dc-use-key="${escHtml(useEntry.settingKey)}"
            data-dc-name="${n}"
-           title="Saves this URL to Settings and makes it available in the ${escHtml(useEntry.label.replace('Use as ', ''))} dropdown. Also applied automatically when you Connect.">
+           title="Makes this URL and model active in the ${escHtml(useLabel.replace('Use as ', ''))} dropdown.">
-           <span class="mdi mdi-arrow-right-circle-outline"></span> ${escHtml(useEntry.label)}
+           <span class="mdi mdi-arrow-right-circle-outline"></span> ${escHtml(useLabel)}
          </button>`
       : '';
 
@@ -586,6 +759,7 @@ function renderLocalContainers(containers) {
     const controlsRowHtml = `
   <div class="dc-controls-row">
     <button class="llm-local-ping dc-connect-btn" data-dc-url-key="${n}"><span class="mdi mdi-lan-connect"></span> Connect</button>
+    <button class="llm-use-btn dc-disconnect-btn" data-dc-name="${n}" type="button" hidden>Disconnect</button>
     <div class="dc-ctrl-group">${dockerBtnsSpan}${githubBtn}${editBtn}${deleteBtn}</div>
     ${useBtn}
   </div>`;
@@ -622,7 +796,7 @@ function renderLocalContainers(containers) {
     <span class="llm-local-compat ${stCls}">${escHtml(stLabel)}</span>
   </div>
   ${metricsHtml}
-  ${c.description ? `<p class="llm-local-desc">${escHtml(c.description)}</p>` : ''}${urlRowHtml}${apiKeyRowHtml}${containerNameRowHtml}${controlsRowHtml}${snippetHtml}
+  ${c.description ? `<p class="llm-local-desc">${escHtml(c.description)}</p>` : ''}${urlRowHtml}${apiKeyRowHtml}${modelRowHtml}${containerNameRowHtml}${controlsRowHtml}${snippetHtml}
 </div>`;
   }
 
@@ -673,32 +847,33 @@ function renderLocalContainers(containers) {
 
     // Helpers — set connected / disconnected visual state on a card
     function setConnected(btn, card, useBtn, connected) {
+      const disconnectBtn = card?.querySelector('.dc-disconnect-btn');
+      if (disconnectBtn) disconnectBtn.hidden = !connected;
       if (connected) {
         btn.innerHTML = '<span class="mdi mdi-check-network"></span> Connected';
         btn.className = 'llm-local-ping dc-connect-btn ok';
-        btn.title = 'Reachable — click to disconnect';
+        btn.title = 'Reachable — click to test again';
         if (card) card.classList.add('llm-local-card-online');
-        if (useBtn) useBtn.classList.add('active');
       } else {
         btn.innerHTML = '<span class="mdi mdi-lan-connect"></span> Connect';
         btn.className = 'llm-local-ping dc-connect-btn';
         btn.title = 'Test connection to this service';
         if (card) card.classList.remove('llm-local-card-online');
-        if (useBtn) useBtn.classList.remove('active');
       }
     }
 
     // URL input — server settings first, localStorage fallback; write to both on change
     grid.querySelectorAll('.dc-url-inp[data-dc-url-key]').forEach(inp => {
       const key       = inp.dataset.dcUrlKey;
-      const fromSrv   = _appSettings?.engine_local_urls?.['dc-' + key];
+      const legacy = inp.dataset.dcLegacyName;
-      const fromLS    = localStorage.getItem('dc-url-' + key);
+      const fromSrv   = _appSettings?.engine_local_urls?.['dc-' + key] || _appSettings?.engine_local_urls?.['dc-' + legacy];
+      const fromLS    = localStorage.getItem('dc-url-' + key) || localStorage.getItem('dc-url-' + legacy);
       if (fromSrv || fromLS) inp.value = fromSrv || fromLS;
       inp.addEventListener('input', () => {
         localStorage.setItem('dc-url-' + key, inp.value);
         if (window._saveEngineLocalUrls) window._saveEngineLocalUrls();
       });
-      if (localStorage.getItem('dc-con-' + key) === '1') {
+      if (_appSettings?.engine_disabled?.[key] !== true && localStorage.getItem('dc-con-' + key) === '1') {
         const card   = inp.closest('.llm-local-card');
         const btn    = card?.querySelector(`.dc-connect-btn[data-dc-url-key="${CSS.escape(key)}"]`);
         const useBtn = card?.querySelector('.dc-use-btn');
@@ -706,6 +881,50 @@ function renderLocalContainers(containers) {
       }
     });
 
+    grid.querySelectorAll('.dc-disconnect-btn').forEach(btn => {
+      btn.addEventListener('click', async () => {
+        btn.disabled = true;
+        try {
+          const key = btn.dataset.dcName;
+          const card = btn.closest('.llm-local-card');
+          await setEngineConnection(key, false, cardType(card));
+          localStorage.removeItem('dc-con-' + key);
+          const useBtn = card.querySelector('.dc-use-btn');
+          setConnected(card.querySelector('.dc-connect-btn'), card, useBtn, false);
+          useBtn?.classList.remove('active');
+          toast('Engine disconnected', 'success');
+        } catch (e) { toast('Disconnect failed: ' + e.message, 'error'); }
+        finally { btn.disabled = false; }
+      });
+    });
+
+    grid.querySelectorAll('.dc-models-btn').forEach(btn => {
+      btn.addEventListener('click', async () => {
+        const card = btn.closest('.llm-local-card');
+        const inp = card.querySelector('.dc-url-inp');
+        const key = inp.dataset.dcUrlKey;
+        const url = inp.value.trim() || inp.placeholder;
+        const apiKey = card.querySelector('.dc-apikey-inp').value.trim();
+        if (!url) { toast('Enter a URL first', 'error'); return; }
+        btn.disabled = true;
+        try {
+          const params = new URLSearchParams({ url, type: cardType(card), name: key });
+          const d = await fetch('/api/engine-models?' + params, { headers: apiKey ? { 'X-Provider-API-Key': apiKey } : {} }).then(r => r.json());
+          if (!d.ok) { toast(d.error || 'Could not load models', 'error'); return; }
+          const models = d.models || [];
+          card.querySelector('datalist').innerHTML = models.map(model => `<option value="${escHtml(model)}"></option>`).join('');
+          toast(models.length ? `Loaded ${models.length} models` : 'No models advertised — enter an ID manually', 'success');
+        } catch (e) { toast('Failed to load models: ' + e.message, 'error'); }
+        finally { btn.disabled = false; }
+      });
+    });
+    grid.querySelectorAll('.dc-model-inp').forEach(inp => {
+      inp.addEventListener('change', async () => {
+        const key = inp.closest('.llm-local-card').querySelector('.dc-url-inp').dataset.dcUrlKey;
+        await window.applyAndSaveSettings({ engine_models: { ...(_appSettings?.engine_models || {}), [key]: inp.value.trim() } });
+      });
+    });
+
     // API key input — save only to server-backed engine_api_keys
     grid.querySelectorAll('.dc-apikey-inp[data-dc-apikey-key]').forEach(inp => {
       const key = inp.dataset.dcApikeyKey;
@@ -725,15 +944,6 @@ function renderLocalContainers(containers) {
     // Connect / Disconnect — probe URL, auto-apply setting, show state
     grid.querySelectorAll('.dc-connect-btn[data-dc-url-key]').forEach(btn => {
       btn.addEventListener('click', async () => {
-        // Disconnect if already connected
-        if (btn.classList.contains('ok')) {
-          const key  = btn.dataset.dcUrlKey;
-          const card = btn.closest('.llm-local-card');
-          const useBtn = card?.querySelector('.dc-use-btn');
-          setConnected(btn, card, useBtn, false);
-          localStorage.removeItem('dc-con-' + key);
-          return;
-        }
         const key    = btn.dataset.dcUrlKey;
         const card   = btn.closest('.llm-local-card');
         const inp    = card?.querySelector(`.dc-url-inp[data-dc-url-key="${CSS.escape(key)}"]`);
@@ -749,18 +959,10 @@ function renderLocalContainers(containers) {
             const type = card ? cardType(card) : '';
             const d    = await window.probeUrl(rawUrl, type, apiKey, key);
             if (d.ok) {
+              await setEngineConnection(key, true, type);
               setConnected(btn, card, useBtn, true);
               localStorage.setItem('dc-con-' + key, '1');
-              // Auto-apply URL to settings so it appears in dropdowns immediately.
+              toast('✓ Service reachable', 'success');
-              // Reuses the "Use as" button's own click handler (rather than
-              // duplicating its logic) so LLM cards go through the same
-              // credential-aware /use-as-llm endpoint Connect just verified.
-              if (useBtn) {
-                useBtn.click();
-                toast(`✓ Connected and set as active ${useBtn.textContent.replace('Use as ', '').trim()}`, 'success');
-              } else {
-                toast('✓ Service reachable', 'success');
-              }
             } else {
               setConnected(btn, card, useBtn, false);
               localStorage.removeItem('dc-con-' + key);
@@ -789,32 +991,17 @@ function renderLocalContainers(containers) {
         const url = inp?.value.trim() || inp?.placeholder || '';
         const apiKey = apikeyInp?.value.trim() || '';
         if (!url) { toast('Enter a URL first', 'error'); return; }
-        if (settingKey === 'llm_url') {
+
-          // Goes through a dedicated endpoint (not applyAndSaveSettings) so the
+        const role = cardType(card);
-          // backend can fall back to this card's own saved key — the browser
+        const model = card?.querySelector('.dc-model-inp')?.value.trim() || '';
-          // only has a real key in hand here if the user just typed one.
+
-          try {
+        try {
-            const headers = { 'Content-Type': 'application/json' };
+          await activateEngine(urlKey, role, settingKey, url, model, apiKey);
-            if (apiKey) headers['X-Provider-API-Key'] = apiKey;
-            const r = await fetch(`/api/local-containers/${encodeURIComponent(urlKey)}/use-as-llm`, {
-              method: 'POST', headers, body: JSON.stringify({ url }),
-            });
-            const d = await r.json();
-            if (!r.ok || !d.ok) { toast(d.error || 'Failed to set active LLM', 'error'); return; }
-            if (_appSettings) { _appSettings.llm_url = url; if (d.key_set) _appSettings.llm_api_key = '__configured__'; }
-            if (window.syncActiveLlmPanel) window.syncActiveLlmPanel(url, apiKey);
-            btn.classList.add('active');
-            toast(`✓ ${btn.textContent.trim()} — now available in the backend dropdown`, 'success');
-          } catch (e) { toast('Failed to set active LLM: ' + e.message, 'error'); }
-          return;
-        }
-        if (window.applyAndSaveSettings) {
-          const patch = { [settingKey]: url };
-          await window.applyAndSaveSettings(patch);
-          if (settingKey === 'vision_ocr_url' && window.syncActiveOcrPanel) window.syncActiveOcrPanel(url);
           btn.classList.add('active');
+          setConnected(card.querySelector('.dc-connect-btn'), card, btn, true);
+          localStorage.setItem('dc-con-' + urlKey, '1');
           toast(`✓ ${btn.textContent.trim()} — now available in the backend dropdown`, 'success');
-        }
+        } catch (e) { toast('Failed to set active engine: ' + e.message, 'error'); }
       });
     });
 

static/js/settings.js

@@ -41,7 +41,7 @@ function ttsBackendOptions(selected = '') {
   const backends = availableTtsBackends();
   if (!backends.length) return '<option value="">No TTS backend available</option>';
   const current = selected || backends[0].id;
-  return backends.map(b => `<option value="${escHtml(b.id)}" ${b.id === current ? 'selected' : ''}>${escHtml(b.label)}</option>`).join('');
+  return backends.map(b => `<option value="${escHtml(b.id)}" ${b.id === current ? 'selected' : ''}>${escHtml(b.title || b.label || b.id)}</option>`).join('');
 }
 
 function styleBackendOptions(selected = 'customvoice', preferStyleAware = false) {
@@ -51,7 +51,7 @@ function styleBackendOptions(selected = 'customvoice', preferStyleAware = false)
     : preferStyleAware
       ? (backends.find(b => b.style_aware)?.id || backends[0].id)
       : backends[0].id;
-  return backends.map(b => `<option value="${escHtml(b.id)}" ${b.id === preferred ? 'selected' : ''}>${escHtml(b.label)}</option>`).join('');
+  return backends.map(b => `<option value="${escHtml(b.id)}" ${b.id === preferred ? 'selected' : ''}>${escHtml(b.title || b.label || b.id)}</option>`).join('');
 }
 
 function backendById(id) {
@@ -178,7 +178,8 @@ function updateBackendDependentTabs() {
     if (tooltip) tooltip.textContent = ok ? originalTooltip : `${originalTooltip}\n\n${disabledBackendTabMessage(tab)}`;
   });
   const active = document.querySelector('.tab.active');
-  if (!active || active.classList.contains('backend-unavailable')) {
+  // Sidebar pages own navigation; their hidden legacy tabs can retain a stale active tab.
+  if (!document.querySelector('.page-section.is-active') && (!active || active.classList.contains('backend-unavailable'))) {
     const first = document.querySelector('.tab:not(.backend-unavailable)');
     if (first) switchTab(first.dataset.tab);
   }
@@ -507,12 +508,37 @@ function _populateImageGenModelSelect(selectEl, provider, savedModel) {
   }
 }
 
+window.syncPreferredSttBackend = function (s) {
+  const select = $('s-stt-preferred-backend');
+  if (!select) return;
+  select.querySelectorAll('[data-custom-engine]').forEach(option => option.remove());
+  for (const card of s.custom_engine_cards || []) {
+    if (card.role !== 'stt' && card.role !== 'stt+tts') continue;
+    const value = 'custom:' + (card.id || card.name);
+    if (Array.from(select.options).some(option => option.value === value)) continue;
+    const option = new Option(card.label || card.name, value);
+    option.dataset.customEngine = '1';
+    select.add(option);
+  }
+  if (s.stt_preferred_backend?.startsWith('custom:') && !Array.from(select.options).some(option => option.value === s.stt_preferred_backend)) {
+    const option = new Option(s.stt_preferred_backend, s.stt_preferred_backend);
+    option.dataset.customEngine = '1';
+    select.add(option);
+  }
+  select.value = s.stt_preferred_backend || '';
+};
+
 async function loadSettings() {
   const s = await fetch('/api/settings').then(r => r.json());
   const sv = (id, val) => { const el = $(id); if (el) el.value = val; };
   const svSecret = (id, val) => { const el = $(id); if (el) { el.value = ''; el.placeholder = val === '__configured__' ? 'Configured — enter a new key to replace' : ''; } };
   sv('s-whisper-url',        s.whisper_url          || '');
   svSecret('s-whisper-key',   s.whisper_api_key);
+  svSecret('s-stt-api-key',   s.whisper_api_key);
+  if ($('s-stt-model')) {
+    const model = s.stt_models?.configured || '';
+    $('s-stt-model').innerHTML = `<option value="${escHtml(model)}">${escHtml(model || '— load models —')}</option>`;
+  }
   sv('s-tts-url',            s.tts_url              || '');
   sv('s-faster-whisper-url', s.faster_whisper_url   || '');
   sv('s-whisper-cpp-url',    s.whisper_cpp_url      || '');
@@ -748,7 +774,7 @@ async function loadSettings() {
   if (window.populateLangSelect) window.populateLangSelect();
   // Captures settings
   const sttLang = $('s-stt-language'); if (sttLang) sttLang.value = s.stt_language || '';
-  const sttPref = $('s-stt-preferred-backend'); if (sttPref) sttPref.value = s.stt_preferred_backend || '';
+  window.syncPreferredSttBackend(s);
   const autoRef = $('s-auto-refine'); if (autoRef) autoRef.value = s.auto_refine || 'off';
   const refModel = $('s-refine-model'); if (refModel) refModel.value = s.refine_model || '';
   const rfFill = $('s-refine-fillers'); if (rfFill) rfFill.checked = s.refine_fillers !== false;
@@ -818,7 +844,8 @@ document.addEventListener('click', async e => { if (!e.target.closest('.s-save-b
   await fetch('/api/settings', { method:'POST', headers:{'Content-Type':'application/json'},
     body: JSON.stringify({
       whisper_url:          $('s-whisper-url').value,
-      whisper_api_key:      $('s-whisper-key').value,
+      whisper_api_key:      $('s-stt-api-key')?.value || $('s-whisper-key').value,
+      stt_models:          { ...(_appSettings?.stt_models || {}), configured: $('s-stt-model')?.value || '' },
       faster_whisper_url:   $('s-faster-whisper-url').value,
       whisper_cpp_url:      $('s-whisper-cpp-url').value,
       groq_api_key:         $('s-groq-api-key').value,

static/js/stt.js

@@ -16,12 +16,13 @@ function sttTtsSelectedSttBackend() {
 }
 
 function sttBackendOptionHtml(selected = 'configured') {
-  if (!_sttBackends.length) return '<option value="configured">Configured Whisper/STT</option>';
+  const choices = sttBackendChoices(_sttBackends, selected);
-  const preferred = _sttBackends.some(b => b.id === selected && b.available) ? selected : (_sttBackends.find(b => b.available)?.id || selected);
+  if (!choices.length) return '<option value="configured">Configured Whisper/STT</option>';
-  return _sttBackends.map(b => {
+  const preferred = choices.some(b => b.id === selected && b.available) ? selected : (choices.find(b => b.available)?.id || selected);
+  return choices.map(b => {
     const suffix = b.available ? '' : ' (unavailable)';
     const disabled = b.available ? '' : ' disabled';
-    return `<option value="${escHtml(b.id)}" ${b.id === preferred ? 'selected' : ''}${disabled}>${escHtml(b.label + suffix)}</option>`;
+    return `<option value="${escHtml(b.id)}" ${b.id === preferred ? 'selected' : ''}${disabled}>${escHtml((b.title || b.label || b.id) + suffix)}</option>`;
   }).join('');
 }
 
@@ -37,14 +38,14 @@ function updateSttBackendHelp() {
 async function refreshSttBackends(selected = '') {
   try {
     const d = await fetch('/api/stt-backends').then(r => r.json());
-    _sttBackends = (d.backends || []).filter(b => b && b.id);
+    _sttBackends = (d.backends || []).filter(b => b && b.id).map(b => b.connected === false ? { ...b, available: false } : b);
   } catch (_) {
     _sttBackends = [];
   }
-  ['stt-tts-stt-backend', 'clone-stt-backend'].forEach(id => {
+  ['conv-stt-select', 'stt-tts-stt-backend', 'clone-stt-backend'].forEach(id => {
     const sel = $(id);
     if (!sel) return;
-    const prev = selected || sel.value || 'configured';
+    const prev = selected || (typeof _appSettings !== 'undefined' && _appSettings?.stt_preferred_backend) || sel.value || 'configured';
     sel.innerHTML = sttBackendOptionHtml(prev);
     sel.disabled = !_sttBackends.some(b => b.available);
   });

static/js/utils.js

@@ -115,7 +115,7 @@ function statusActiveTts() {
   try {
     all = (typeof availableTtsBackends === 'function') ? availableTtsBackends() : [];
   } catch (_) {}
-  const labels = all.map(b => b.label || b.id);
+  const labels = all.map(b => b.title || b.label || b.id);
   return {
     kind: 'tts',
     label: 'TTS',
@@ -125,21 +125,33 @@ function statusActiveTts() {
   };
 }
 
+function sttBackendChoices(backends, selected = '') {
+  const all = backends.filter(b => b && b.id).map(b => b.connected === false ? { ...b, available: false } : b);
+  const winners = new Map();
+  for (const b of all) {
+    if (!b.available || !b.engine_id) continue;
+    const previous = winners.get(b.engine_id);
+    if (!previous || b.id === selected || (previous.id !== selected && b.id === `custom:${b.engine_id}`)) winners.set(b.engine_id, b);
+  }
+  return all.filter(b => !b.available || !b.engine_id || winners.get(b.engine_id) === b);
+}
+
 function statusActiveStt() {
   let all = [];
   try {
     all = Array.isArray(_sttBackends) ? _sttBackends : [];
   } catch (_) {}
-  const preferred = $('conv-stt-select')?.value || $('stt-tts-stt-backend')?.value || $('clone-stt-backend')?.value
+  const preferred = (typeof _appSettings !== 'undefined' && _appSettings?.stt_preferred_backend)
-    || (typeof _appSettings !== 'undefined' && _appSettings?.stt_preferred_backend) || '';
+    || $('conv-stt-select')?.value || $('stt-tts-stt-backend')?.value || $('clone-stt-backend')?.value || '';
+  all = sttBackendChoices(all, preferred);
   const available = all.filter(b => b.available);
   const chosen = (preferred && all.find(b => b.id === preferred)) || available[0] || all[0] || null;
   return {
     kind: 'stt',
     label: 'STT',
     ok: !!(chosen && chosen.available),
-    value: chosen ? (chosen.label || chosen.model || chosen.id) : 'No STT backend',
+    value: chosen ? (chosen.title || chosen.label || chosen.id) : 'No STT backend',
-    title: chosen ? `${chosen.label || chosen.id} · ${chosen.url || ''}` : 'No STT backend configured',
+    title: chosen ? `${chosen.title || chosen.label || chosen.id} · ${chosen.url || ''}` : 'No STT backend configured',
   };
 }
 
@@ -149,6 +161,9 @@ function statusActiveStt() {
 function statusLlmTarget() {
   let settings = {};
   try { settings = (typeof _appSettings !== 'undefined' && _appSettings) ? _appSettings : {}; } catch (_) {}
+  if (settings.engine_disabled?.[settings.engine_active?.llm_url] === true) {
+    return { url: '', model: '', apiKey: '', disconnected: true };
+  }
   return {
     url: statusCleanUrl(settings.llm_url || $('reh-llm-url')?.value || $('conv-llm-url')?.value || ''),
     model: settings.llm_model || $('reh-llm-model')?.value || $('conv-llm-model-select')?.value || '',
@@ -157,14 +172,14 @@ function statusLlmTarget() {
 }
 
 function statusActiveLlm() {
-  const { url, model } = statusLlmTarget();
+  const { url, model, disconnected } = statusLlmTarget();
   const same = STATUS_LLM_CACHE.url === url && STATUS_LLM_CACHE.model === model;
   const ok = !!url && same && STATUS_LLM_CACHE.ok;
   return {
     kind: 'llm',
     label: 'LLM',
     ok,
-    value: model || statusShortUrl(url) || 'No LLM configured',
+    value: disconnected ? 'Disconnected' : model || statusShortUrl(url) || 'No LLM configured',
     title: url ? `${model || 'default model'} · ${url}` : 'No active LLM endpoint configured',
   };
 }
@@ -247,8 +262,8 @@ function _statusApplyToSelects(ids, value) {
 }
 
 async function _statusOpenLlmFlyup(anchor) {
-  const { url, model } = statusLlmTarget();
+  const { url, model, disconnected } = statusLlmTarget();
-  if (!url) { _statusFlyupList(anchor, [], null, 'No LLM endpoint configured', 'llm'); return; }
+  if (!url) { _statusFlyupList(anchor, [], null, disconnected ? 'LLM disconnected — use Connect to restore' : 'No LLM endpoint configured', 'llm'); return; }
   _statusFlyupList(anchor, [], null, 'Loading models…', 'llm');
   let models = [];
   try {
@@ -257,10 +272,13 @@ async function _statusOpenLlmFlyup(anchor) {
     models = Array.isArray(data.models) ? data.models : [];
   } catch (_) { logErr('status-llm-models', _); }
   if (!_statusFlyup || _statusFlyup.dataset.forKind !== 'llm') return; // closed/switched while awaiting
+  if (statusLlmTarget().disconnected) {
+    _statusFlyupList(anchor, [], null, 'LLM disconnected — use Connect to restore', 'llm'); return;
+  }
   _statusFlyupList(anchor, models.map(m => ({ id: m, label: m, active: m === model })), (picked) => {
     if (typeof _appSettings !== 'undefined') _appSettings.llm_model = picked;
     fetch('/api/settings', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ llm_model: picked }) }).catch(() => {});
-    _statusApplyToSelects(['reh-llm-model', 'conv-llm-model-select'], picked);
+    _statusApplyToSelects(['llm-active-model', 'reh-llm-model', 'conv-llm-model-select'], picked);
     refreshStatusBarEngines({ force: true });
     toast('LLM model set to ' + picked, 'success');
   }, 'No models found at this endpoint', 'llm');
@@ -268,10 +286,10 @@ async function _statusOpenLlmFlyup(anchor) {
 
 function _statusOpenSttFlyup(anchor) {
   const all = Array.isArray(_sttBackends) ? _sttBackends : [];
-  const preferred = $('conv-stt-select')?.value || $('stt-tts-stt-backend')?.value || $('clone-stt-backend')?.value
+  const preferred = (typeof _appSettings !== 'undefined' && _appSettings?.stt_preferred_backend)
-    || (typeof _appSettings !== 'undefined' && _appSettings?.stt_preferred_backend) || '';
+    || $('conv-stt-select')?.value || $('stt-tts-stt-backend')?.value || $('clone-stt-backend')?.value || '';
-  _statusFlyupList(anchor, all.filter(b => b.available).map(b => ({
+  _statusFlyupList(anchor, sttBackendChoices(all, preferred).filter(b => b.available).map(b => ({
-    id: b.id, label: b.label || b.id, active: b.id === preferred,
+    id: b.id, label: b.title || b.label || b.id, active: b.id === preferred,
   })), (picked) => {
     // Persisting this (not just pushing it into a few unrelated <select>s on
     // other pages) is what makes the pick actually take effect anywhere that
@@ -282,6 +300,7 @@ function _statusOpenSttFlyup(anchor) {
     if (typeof _appSettings !== 'undefined') _appSettings.stt_preferred_backend = picked;
     fetch('/api/settings', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ stt_preferred_backend: picked }) }).catch(() => {});
     _statusApplyToSelects(['conv-stt-select', 'stt-tts-stt-backend', 'clone-stt-backend'], picked);
+    window.syncPreferredSttBackend?.(_appSettings);
     updateStatusBar();
     toast('STT backend set to ' + picked, 'success');
   }, 'No reachable STT backend', 'stt');
@@ -289,13 +308,14 @@ function _statusOpenSttFlyup(anchor) {
 
 function _statusOpenTtsFlyup(anchor) {
   const all = (typeof availableTtsBackends === 'function') ? availableTtsBackends() : [];
-  const preferred = $('tts-backend-select')?.value || $('reader-backend-select')?.value || $('reh-backend-select')?.value || '';
+  const preferred = $('library-tts-backend-select')?.value || $('tts-backend-select')?.value || $('reader-backend-select')?.value || $('reh-backend-select')?.value || '';
   _statusFlyupList(anchor, all.map(b => ({
-    id: b.id, label: b.label || b.id, active: b.id === preferred,
+    id: b.id, label: b.title || b.label || b.id, active: b.id === preferred,
   })), (picked) => {
-    _statusApplyToSelects(['tts-backend-select', 'reader-backend-select', 'reh-backend-select'], picked);
+    _statusApplyToSelects(['tts-backend-select', 'library-tts-backend-select', 'stt-tts-backend-select', 'reader-backend-select', 'reh-backend-select'], picked);
     updateStatusBar();
-    toast('TTS backend set to ' + picked, 'success');
+    const chosen = all.find(b => b.id === picked);
+    toast('TTS backend set to ' + ((chosen && (chosen.title || chosen.label)) || picked), 'success');
   }, 'No reachable TTS backend', 'tts');
 }
 
@@ -372,18 +392,49 @@ async function runPool(items, worker, concurrency, onProgress) {
   await Promise.all(Array.from({ length: Math.min(effectiveConcurrency, arr.length) }, run));
 }
 window.runPool = runPool;
+// Resolves true only when the async Clipboard API confirmed the write. The legacy
+// execCommand fallback is still attempted but cannot be verified (it reports
+// success in embedded webviews that discard the copy), so it resolves false and
+// callers show the text via copyTextDialog().
 async function copyText(text) {
   try {
     await navigator.clipboard.writeText(text);
+    return true;
   } catch(e) {
     const ta = document.createElement('textarea');
     ta.value = text;
+    ta.setAttribute('readonly', '');
+    ta.style.cssText = 'position:fixed;top:0;left:0;opacity:0';
     document.body.appendChild(ta);
     ta.select();
-    document.execCommand('copy');
+    let ok = false;
+    try { document.execCommand('copy'); } catch (_) {}
     ta.remove();
+    return false;
   }
 }
+// Last-resort manual copy: the text in a pre-selected box the user can Ctrl+C.
+function copyTextDialog(text, title = 'Copy to clipboard') {
+  const ov = document.createElement('div');
+  ov.className = 'audiobook-overlay';
+  const uid = 'ct-' + Math.random().toString(36).slice(2, 9);
+  ov.innerHTML = `<div class="audiobook-box" role="dialog" aria-modal="true" aria-labelledby="${uid}-t">
+    <div class="audiobook-title" id="${uid}-t"><span class="mdi mdi-content-copy" aria-hidden="true"></span> ${escHtml(title)}</div>
+    <div class="audiobook-msg">The browser blocked automatic copying. Press Ctrl+C (Cmd+C) to copy:</div>
+    <textarea readonly rows="5" style="width:100%" aria-label="Text to copy"></textarea>
+    <div class="audiobook-actions"><button type="button" class="btn-primary btn-sm">Close</button></div>
+  </div>`;
+  const ta = ov.querySelector('textarea');
+  ta.value = text;
+  const close = () => { ov.remove(); document.removeEventListener('keydown', onKey); };
+  const onKey = e => { if (e.key === 'Escape') close(); };
+  document.addEventListener('keydown', onKey);
+  ov.querySelector('button').addEventListener('click', close);
+  ov.addEventListener('click', e => { if (e.target === ov) close(); });
+  document.body.appendChild(ov);
+  ta.focus();
+  ta.select();
+}
 function fmtDuration(sec) {
   if (sec == null || Number.isNaN(Number(sec))) return '-';
   const total = Math.max(0, Math.round(Number(sec)));
@@ -1432,8 +1483,8 @@ else initLangSelects(document);
 })();
 
 Object.assign(root, {
-  $, avatarColor, toast, confirmDialog, status, ensureStatusBar, updateStatusBar,
+  $, avatarColor, toast, confirmDialog, status, ensureStatusBar, updateStatusBar, sttBackendChoices,
-  escHtml, debounce, logErr, ttvcIsLowMemoryDevice, fmtDuration, loadingMarkup,
+  escHtml, copyText, copyTextDialog, debounce, logErr, ttvcIsLowMemoryDevice, fmtDuration, loadingMarkup,
   setBusyButton, enhanceAccessibility, applyTheme, cc2flag, ccDisplay,
   setPreviewLang, openPreviewLangPicker, createSearchablePicker, getStoredTags,
   addStoredTag, scheduleVoiceLibraryLoad, disabledBackendTabMessage, switchTab,

static/js/voice-library.js

@@ -164,10 +164,13 @@ function toggleSortDir() {
 }
 
 function syncSortHeaders() {
-  document.querySelectorAll('.vl-header [data-sort], .vl-table-header [data-sort]').forEach(el => {
+  document.querySelectorAll('.vl-header [data-sort], .vl-table-header [data-sort], .vl-compact-header [data-sort]').forEach(el => {
     el.classList.remove('sort-asc', 'sort-desc');
-    if (el.dataset.sort === _sortField)
+    el.removeAttribute('aria-sort');
+    if (el.dataset.sort === _sortField) {
       el.classList.add(_sortDir === 1 ? 'sort-asc' : 'sort-desc');
+      el.setAttribute('aria-sort', _sortDir === 1 ? 'ascending' : 'descending');
+    }
   });
   const sel = document.getElementById('voice-sort-field');
   if (sel && sel.value !== _sortField) sel.value = _sortField;
@@ -1119,11 +1122,16 @@ $('benchmark-confirm-start')?.addEventListener('click', async () => {
 $('copy-active-voices-btn').addEventListener('click', async () => {
   const useSelected = _bulkSelected && _bulkSelected.size > 0;
   const ids = useSelected ? [..._bulkSelected] : activeVoiceIds();
-  const label = useSelected ? `${ids.length} selected` : `${ids.length} active`;
   if (!ids.length) { toast('No voices to copy', 'error'); return; }
-  await copyText(ids.join(', '));
+  const text = ids.join(', ');
-  toast('Copied ' + label + ' voices', 'success');
+  if (await copyText(text)) {
-  status('Copied ' + label + ' voices to clipboard');
+    const msg = `${ids.length} voice name${ids.length === 1 ? '' : 's'} copied to clipboard`;
+    toast(msg, 'success');
+    status(msg);
+  } else {
+    // Async clipboard unavailable (plain-HTTP LAN origin, embedded webview): show the list to copy by hand.
+    copyTextDialog(text, `${ids.length} voice name${ids.length === 1 ? '' : 's'}`);
+  }
 });
 
 // Round-trip verify: synthesize each voice's own reference line (or a
@@ -3456,15 +3464,22 @@ function renderVoiceGroupsBar() {
   if (!names.length) { bar.hidden = true; bar.innerHTML = ''; return; }
   bar.hidden = false;
   const active = window._voiceGroupFilter || '';
+  // Collapsed by default (accordion): the chip list is long and duplicates the
+  // "All groups" dropdown; the summary line shows the count and the active group.
   bar.innerHTML =
-    `<span class="vg-label"><span class="mdi mdi-folder-multiple-outline"></span> Groups</span>` +
+    `<details class="vg-acc"${window._vgOpen ? ' open' : ''}>
+       <summary class="vg-label"><span class="mdi mdi-folder-multiple-outline" aria-hidden="true"></span> Groups
+         <span class="vg-count">${names.length}</span>
+         ${active ? `<span class="vg-active">${escHtml(active)}</span><button type="button" class="vg-clear" title="Clear group filter">Show all</button>` : ''}
+       </summary>
+       <div class="vg-chips">` +
     names.map(g =>
       `<span class="vg-chip${g===active?' active':''}" data-group="${escHtml(g)}">
          ${escHtml(g)} <span class="vg-count">${groups[g]}</span>
          <button class="vg-del" data-group="${escHtml(g)}" title="Move all ${groups[g]} voices in this group to Voice Trash">&times;</button>
        </span>`
-    ).join('') +
+    ).join('') + `</div></details>`;
-    (active ? `<button class="vg-clear" title="Clear group filter">Show all</button>` : '');
+  bar.querySelector('details').addEventListener('toggle', e => { window._vgOpen = e.target.open; });
 
   bar.querySelectorAll('.vg-chip').forEach(chip => {
     chip.addEventListener('click', e => {
@@ -3478,7 +3493,8 @@ function renderVoiceGroupsBar() {
   bar.querySelectorAll('.vg-del').forEach(btn => {
     btn.addEventListener('click', e => { e.stopPropagation(); deleteVoiceGroup(btn.dataset.group); });
   });
-  bar.querySelector('.vg-clear')?.addEventListener('click', () => {
+  bar.querySelector('.vg-clear')?.addEventListener('click', e => {
+    e.preventDefault(); e.stopPropagation();
     window._voiceGroupFilter = '';
     const groupSel = $('library-filter-group'); if (groupSel) groupSel.value = '';
     renderVoiceList();
@@ -3528,18 +3544,25 @@ function _bulkUpdate() {
   const count = $('vl-bulk-count');
   const n = _bulkSelected.size;
   if (bar)   bar.hidden   = n === 0;
+  const ab = $('vl-actions-btn');
+  if (ab) {
+    ab.disabled = n === 0;
+    ab.title = n === 0 ? 'Check one or more voices to enable actions' : `Actions on ${n} checked voice${n === 1 ? '' : 's'}`;
+    const cnt = $('vl-actions-n'); if (cnt) cnt.textContent = n ? `(${n})` : '';
+    if (n === 0) _setActionsMenu(false);
+  }
   if (count) count.textContent = `${n} selected`;
   // Sync checkboxes in the DOM to match the set
   document.querySelectorAll('.vl-bulk-cb').forEach(cb => {
     cb.checked = _bulkSelected.has(cb.dataset.id);
   });
   
-  const headerCb = document.getElementById('vl-select-all-header');
+  const rows = [...document.querySelectorAll('#voice-list .vl-row')].filter(r => r.dataset.id);
-  if (headerCb) {
+  const nSel = rows.filter(r => _bulkSelected.has(r.dataset.id)).length;
-    const rows = [...document.querySelectorAll('#voice-list .vl-row')].filter(r => r.dataset.id);
+  document.querySelectorAll('.vl-select-all-cb').forEach(cb => {
-    const allSelected = rows.length > 0 && rows.every(r => _bulkSelected.has(r.dataset.id));
+    cb.checked = rows.length > 0 && nSel === rows.length;
-    headerCb.checked = allSelected;
+    cb.indeterminate = nSel > 0 && nSel < rows.length;
-  }
+  });
 }
 
 function _bulkToggle(id, checked) {
@@ -3552,8 +3575,69 @@ const _origRenderVoiceList = renderVoiceList;
 renderVoiceList = function() {
   _origRenderVoiceList.apply(this, arguments);
   _bulkInjectCheckboxes();
+  _bulkUpdate();
+  _updateFilterBadge();
 };
 
+// Actions menu (bottom bar): one button instead of a dozen; closes on pick, outside click or Esc.
+function _setActionsMenu(open) {
+  const m = $('vl-actions-menu'), b = $('vl-actions-btn');
+  if (!m || !b) return;
+  m.hidden = !open;
+  b.setAttribute('aria-expanded', String(open));
+}
+$('vl-actions-btn')?.addEventListener('click', () => _setActionsMenu($('vl-actions-menu').hidden));
+$('vl-actions-menu')?.addEventListener('click', e => { if (e.target.closest('button')) _setActionsMenu(false); });
+document.addEventListener('click', e => { if (!e.target.closest('.vl-actions')) _setActionsMenu(false); });
+document.addEventListener('keydown', e => {
+  if (e.key === 'Escape' && !$('vl-actions-menu')?.hidden) { _setActionsMenu(false); $('vl-actions-btn')?.focus(); }
+});
+
+// Preview-text / synth-source drawer: an icon in the top bar instead of a permanent card.
+function _synthModeLabel() {
+  return document.querySelector('#vl-synth-mode-seg .vl-synth-seg-btn.active')?.textContent.trim() || 'Preview text';
+}
+function _refreshSynthToggle() {
+  const t = $('vl-synth-toggle');
+  if (t) t.title = `Synth uses: ${_synthModeLabel()} — click to ${$('vl-synth-drawer')?.hidden ? 'edit' : 'hide'}`;
+}
+function _setSynthDrawer(open) {
+  const d = $('vl-synth-drawer'), t = $('vl-synth-toggle');
+  if (!d || !t) return;
+  d.hidden = !open;
+  t.setAttribute('aria-expanded', String(open));
+  t.classList.toggle('active', open);
+  _refreshSynthToggle();
+  try { localStorage.setItem('vl-synth-open', open ? '1' : '0'); } catch (_) {}
+}
+$('vl-synth-toggle')?.addEventListener('click', () => _setSynthDrawer($('vl-synth-drawer').hidden));
+document.addEventListener('click', e => { if (e.target.closest('#vl-synth-mode-seg')) setTimeout(_refreshSynthToggle, 0); });
+try { _setSynthDrawer(localStorage.getItem('vl-synth-open') === '1'); } catch (_) {}
+
+// Filter panel: collapsed by default; the badge counts the filters that are narrowing the list.
+function _updateFilterBadge() {
+  const badge = $('vl-filter-badge');
+  if (!badge) return;
+  const val = id => ($(id)?.value || '').trim();
+  const n = ['library-filter-lang', 'library-filter-sex', 'library-filter-type', 'library-filter-tag',
+    'library-filter-rating', 'library-filter-group'].filter(id => val(id)).length
+    + ($('show-disabled-cb')?.checked ? 1 : 0) + (val('library-filter-text') ? 1 : 0);
+  badge.textContent = String(n);
+  badge.hidden = n === 0;
+  const t = $('vl-filter-toggle');
+  if (t) t.title = n ? `Filters (${n} active)` : 'Show filters and sorting';
+}
+function _setFilterPanel(open) {
+  const panel = $('vl-filter-panel'), t = $('vl-filter-toggle');
+  if (!panel || !t) return;
+  panel.hidden = !open;
+  t.setAttribute('aria-expanded', String(open));
+  t.classList.toggle('active', open);
+  try { localStorage.setItem('vl-filter-open', open ? '1' : '0'); } catch (_) {}
+}
+$('vl-filter-toggle')?.addEventListener('click', () => _setFilterPanel($('vl-filter-panel').hidden));
+try { _setFilterPanel(localStorage.getItem('vl-filter-open') === '1'); } catch (_) {}
+
 // Shift-click range select: remembers the last checkbox actually clicked
 // (not just toggled programmatically) so a shift-click on a later row can
 // fill in everything between them — the same convention as a file manager
@@ -3598,17 +3682,11 @@ function _bulkInjectCheckboxes() {
 
 // ── Toolbar button wiring ───────────────────────────────────────────────────
 
-$('vl-bulk-select-all')?.addEventListener('click', () => {
-  document.querySelectorAll('#voice-list .vl-row').forEach(row => {
-    if (row.dataset.id) _bulkSelected.add(row.dataset.id);
-  });
-  _bulkUpdate();
-});
-
 // Always-visible Select-all toggle in the table header (the bulk bar only appears once
 // something is selected, so this is the entry point). Selects every visible row, or
 // clears the selection if they're all already selected.
-$('vl-select-all-header')?.addEventListener('change', (e) => {
+document.addEventListener('change', (e) => {
+  if (!e.target.classList.contains('vl-select-all-cb')) return;
   const rows = [...document.querySelectorAll('#voice-list .vl-row')].filter(r => r.dataset.id);
   if (e.target.checked) {
     rows.forEach(r => _bulkSelected.add(r.dataset.id));
@@ -3618,11 +3696,6 @@ $('vl-select-all-header')?.addEventListener('change', (e) => {
   _bulkUpdate();
 });
 
-$('vl-bulk-deselect')?.addEventListener('click', () => {
-  _bulkSelected.clear();
-  _bulkUpdate();
-});
-
 $('vl-bulk-hide')?.addEventListener('click', async () => {
   if (!_bulkSelected.size) return;
   const ids = [..._bulkSelected];

static/sections/s-llms.html

@@ -431,6 +431,15 @@
               <label>Active STT URL</label>
               <input type="text" id="s-whisper-url" placeholder="http://host.docker.internal:8010">
               <span class="s-hint">Default recognition endpoint. Expected: <code>POST /v1/audio/transcriptions</code>.</span>
+              <label for="s-stt-api-key">API key</label>
+              <input type="password" id="s-stt-api-key" autocomplete="off" placeholder="API key (optional)">
+              <label for="s-stt-model">Model</label>
+              <select id="s-stt-model"><option value="">— load models —</option></select>
+              <div class="btn-row settings-mini-actions">
+                <button class="btn-secondary" id="s-stt-test-connection" type="button">Test connection</button>
+                <button class="btn-secondary" id="s-stt-load-models" type="button">Load models</button>
+                <span id="s-stt-connection-status" role="status"></span>
+              </div>
               <div class="btn-row settings-mini-actions">
                 <button class="btn-secondary" id="s-use-parakeet-asr" type="button">Use Parakeet</button>
                 <button class="btn-secondary" id="s-use-nvidia-router" type="button">Use router</button>
@@ -844,7 +853,7 @@
       </div><!-- /ocr -->
 
       <!-- ── Add Custom Engine dialog ─────────────────────────────── -->
-      <dialog id="add-engine-dlg" class="add-engine-dlg">
+      <dialog id="add-engine-dlg" class="add-engine-dlg" style="max-height:90dvh;overflow-y:auto">
         <div class="add-engine-form">
           <h3 class="add-engine-title"><span class="mdi mdi-plus-circle-outline"></span> Add Custom Engine</h3>
           <div class="add-engine-field">
@@ -864,6 +873,28 @@
             <label for="aed-url">URL</label>
             <input id="aed-url" type="text" placeholder="http://192.168.1.100:8080" spellcheck="false">
           </div>
+          <div class="add-engine-field">
+            <label for="aed-api-key">API key <span class="add-engine-opt">(optional)</span></label>
+            <input id="aed-api-key" type="password" autocomplete="off" placeholder="API key (optional)">
+          </div>
+          <div class="add-engine-field">
+            <label for="aed-model">Model <span class="add-engine-opt">(optional — select or enter an ID)</span></label>
+            <input id="aed-model" type="text" list="aed-models" placeholder="Model ID" spellcheck="false">
+            <datalist id="aed-models"></datalist>
+            <div class="btn-row settings-mini-actions">
+              <button id="aed-test" type="button" class="btn-secondary">Test connection</button>
+              <button id="aed-load-models" type="button" class="btn-secondary">Load models</button>
+            </div>
+            <span id="aed-status" role="status"></span>
+          </div>
+          <div class="add-engine-field">
+            <label for="aed-params">Additional parameters (JSON)</label>
+            <textarea id="aed-params" rows="3" spellcheck="false" placeholder='{"temperature":0.7}'></textarea>
+            <span class="add-engine-opt">Optional request parameters. Keep credentials in the API key field.</span>
+          </div>
+          <div class="add-engine-field">
+            <label><input id="aed-active" type="checkbox"> Use connection actively after saving</label>
+          </div>
           <div class="add-engine-field">
             <label for="aed-container">Docker container name <span class="add-engine-opt">(optional — enables Stop / Start / Restart)</span></label>
             <input id="aed-container" type="text" placeholder="my-tts-container" spellcheck="false">

static/sections/s-voices.html

@@ -17,9 +17,19 @@
         <!-- LEFT: compact voice list — two cards -->
         <div class="voices-list-pane">
 
-          <!-- Card 1: TTS synth mode + preview sentence -->
+          <!-- Card 2: filters + sort + voice list + toolbar -->
-          <div class="vl-card vl-card-synth">
+          <div class="vl-card vl-card-list">
-            <div class="vl-synth-panel">
+            <!-- Top bar: search + filter toggle + view buttons. Filters/sort live in a collapsible panel. -->
+            <div class="vl-topbar">
+              <input type="search" id="library-filter-text" placeholder="Search voices&#x2026;" aria-label="Search voices" autocomplete="off" class="vl-search vl-search-main">
+              <button id="vl-synth-toggle" class="vl-sort-dir-btn" type="button" aria-expanded="false" aria-controls="vl-synth-drawer" title="Preview text &amp; synth source" aria-label="Preview text and synth source"><span class="mdi mdi-account-voice" aria-hidden="true"></span></button>
+              <button id="vl-filter-toggle" class="vl-sort-dir-btn vl-filter-toggle" type="button" aria-expanded="false" aria-controls="vl-filter-panel" title="Show filters and sorting"><span class="mdi mdi-filter-variant" aria-hidden="true"></span><span class="vl-filter-badge" id="vl-filter-badge" hidden>0</span><span class="sr-only"> Filters</span></button>
+              <button id="voice-group-tag-btn" class="vl-sort-dir-btn" type="button" title="Group into virtual folders by tag" aria-label="Group by tag"><span class="mdi mdi-folder-outline" aria-hidden="true"></span></button>
+              <button id="voice-table-view-btn" class="vl-sort-dir-btn" type="button" title="Toggle full-width table view" aria-label="Table view"><span class="mdi mdi-table" aria-hidden="true"></span></button>
+              <button id="voice-table-edit-btn" class="vl-sort-dir-btn" type="button" title="Toggle table edit mode" aria-label="Table edit mode" style="display:none"><span class="mdi mdi-pencil-box-outline" aria-hidden="true"></span></button>
+            </div>
+            <div class="vl-synth-drawer" id="vl-synth-drawer" hidden>
+              <div class="vl-synth-panel">
               <div class="vl-synth-mode-row">
                 <span class="vl-synth-icon"><span class="mdi mdi-play"></span><span class="mdi mdi-play"></span></span>
                 <span class="vl-synth-label">Synth uses</span>
@@ -44,17 +54,15 @@
                   ></textarea>
                 </div>
               </div>
+              </div>
             </div>
-          </div><!-- /vl-card-synth -->
+            <div class="vl-filter-panel" id="vl-filter-panel" role="group" aria-label="Filter and sort voices" hidden>
-
+              <div class="vl-filter-grid">
-          <!-- Card 2: filters + sort + voice list + toolbar -->
+                <label>Language <select id="library-filter-lang"><option value="">All</option></select></label>
-          <div class="vl-card vl-card-list">
+                <label>Gender <select id="library-filter-sex"><option value="">All</option></select></label>
-            <div class="vl-filters">
+                <label>Type <select id="library-filter-type"><option value="">All types</option></select></label>
-              <select id="library-filter-lang" title="Language"><option value="">All</option></select>
+                <label>Tag <select id="library-filter-tag"><option value="">All tags</option></select></label>
-              <select id="library-filter-sex" title="Gender"><option value="">All</option></select>
+                <label>Rating <select id="library-filter-rating" >
-              <select id="library-filter-type" title="Type"><option value="">All types</option></select>
-              <select id="library-filter-tag" title="Tag"><option value="">All tags</option></select>
-              <select id="library-filter-rating" title="Rating">
                 <option value="">All ratings</option>
                 <option value="5">★★★★★</option>
                 <option value="4">★★★★+</option>
@@ -62,15 +70,9 @@
                 <option value="2">★★+</option>
                 <option value="1">★+</option>
                 <option value="0">Unrated</option>
-              </select>
+              </select></label>
-              <select id="library-filter-group" title="Group"><option value="">All groups</option></select>
+                <label>Group <select id="library-filter-group"><option value="">All groups</option></select></label>
-              <button id="library-clear-filters" class="vl-sort-dir-btn" title="Clear all filters"><span class="mdi mdi-filter-off-outline"></span></button>
+                <label>Sort by <select id="voice-sort-field" title="Sort field">
-              <label class="vl-disabled-label"><input type="checkbox" id="show-disabled-cb"> Disabled</label>
-            </div>
-            <div class="vl-sort-bar">
-              <input type="search" id="library-filter-text" placeholder="&#x1F50D; Search voices&#x2026;" autocomplete="off" class="vl-search vl-search-main">
-              <span class="vl-sort-label">Sort</span>
-              <select id="voice-sort-field" title="Sort field">
                 <option value="id">Name</option>
                 <option value="flag">Language</option>
                 <option value="gender">Gender</option>
@@ -83,11 +85,14 @@
                 <option value="note">Note</option>
                 <option value="tag">Tags</option>
                 <option value="enabled">Active</option>
-              </select>
+              </select></label>
-              <button id="voice-sort-dir" class="vl-sort-dir-btn" title="Toggle sort direction"><span class="mdi mdi-arrow-up"></span></button>
+                <label>Order <button id="voice-sort-dir" class="vl-sort-dir-btn vl-sort-dir-wide" type="button" title="Toggle sort direction"><span class="mdi mdi-arrow-up" aria-hidden="true"></span></button></label>
-              <button id="voice-group-tag-btn" class="vl-sort-dir-btn" title="Group into virtual folders by tag"><span class="mdi mdi-folder-outline"></span></button>
+              </div>
-              <button id="voice-table-view-btn" class="vl-sort-dir-btn" title="Toggle full-width table view"><span class="mdi mdi-table"></span></button>
+              <div class="vl-filter-actions">
-              <button id="voice-table-edit-btn" class="vl-sort-dir-btn" title="Toggle table edit mode" style="display:none"><span class="mdi mdi-pencil-box-outline"></span></button>
+                <label class="vl-disabled-label"><input type="checkbox" id="show-disabled-cb"> Show disabled</label>
+                <button id="library-clear-filters" class="btn-secondary vl-tb-btn" type="button" title="Clear all filters"><span class="mdi mdi-filter-off-outline" aria-hidden="true"></span> Clear filters</button>
+              </div>
+              <div id="voice-groups-bar" class="voice-groups-bar" hidden></div>
             </div>
 
             <div class="benchmark-confirm" id="benchmark-confirm" hidden role="group" aria-live="polite">
@@ -97,25 +102,14 @@
               <button class="benchmark-confirm-start" id="benchmark-confirm-start" type="button">Start benchmark</button>
             </div>
 
-            <div id="voice-groups-bar" class="voice-groups-bar" hidden></div>
 
             <!-- Bulk-edit toolbar — appears when voices are selected -->
             <div class="vl-bulk-bar" id="vl-bulk-bar" hidden>
-              <span class="vl-bulk-count" id="vl-bulk-count">0 selected</span>
+              <span class="vl-bulk-count" id="vl-bulk-count" role="status">0 selected</span>
-              <button class="vl-bulk-btn" id="vl-bulk-select-all" title="Select all visible"><span class="mdi mdi-checkbox-multiple-marked-outline"></span> All</button>
-              <button class="vl-bulk-btn" id="vl-bulk-deselect" title="Clear selection"><span class="mdi mdi-checkbox-multiple-blank-outline"></span> None</button>
-              <div class="vl-bulk-sep"></div>
-              <button class="vl-bulk-btn" id="vl-bulk-tag" title="Add a tag to the selected voices, keeping their existing tags"><span class="mdi mdi-tag-outline"></span> Set tag</button>
-              <button class="vl-bulk-btn" id="vl-bulk-source" title="Set source / origin on selected voices"><span class="mdi mdi-database-outline"></span> Set source</button>
-              <button class="vl-bulk-btn" id="vl-bulk-hide" title="Disable (hide) selected voices"><span class="mdi mdi-eye-off-outline"></span> Hide</button>
-              <button class="vl-bulk-btn" id="vl-bulk-unhide" title="Enable (unhide) selected voices"><span class="mdi mdi-eye-outline"></span> Unhide</button>
-              <button class="vl-bulk-btn" id="vl-bulk-rating" title="Set star rating on selected"><span class="mdi mdi-star-outline"></span> Rate</button>
-              <div class="vl-bulk-sep"></div>
-              <button class="vl-bulk-btn vl-bulk-danger" id="vl-bulk-delete" title="Delete selected voices permanently"><span class="mdi mdi-delete-outline"></span> Delete</button>
             </div>
 
             <div class="vl-table-header" id="vl-table-header">
-              <div style="display:flex; justify-content:center; align-items:center;"><input type="checkbox" id="vl-select-all-header" title="Select all visible voices"></div>
+              <div style="display:flex; justify-content:center; align-items:center;"><input type="checkbox" id="vl-select-all-header" class="vl-select-all-cb" title="Select all visible voices" aria-label="Select all visible voices"></div>
               <div>Img</div>
               <div>Play</div>
               <div class="vl-th-sortable" data-sort="id">Name <span class="mdi mdi-sort"></span></div>
@@ -132,19 +126,42 @@
               <div class="vl-th-sortable" data-sort="note">Note <span class="mdi mdi-sort"></span></div>
               <div class="vl-th-sortable" data-sort="enabled">Active <span class="mdi mdi-sort"></span></div>
             </div>
+            <div class="vl-compact-header" id="vl-compact-header" role="group" aria-label="Sort voices">
+              <input type="checkbox" class="vl-select-all-cb" title="Select all / none of the visible voices" aria-label="Select all visible voices">
+              <button type="button" class="vl-th-sortable" data-sort="id">Name <span class="mdi mdi-sort" aria-hidden="true"></span></button>
+              <button type="button" class="vl-th-sortable" data-sort="flag">Lang <span class="mdi mdi-sort" aria-hidden="true"></span></button>
+              <button type="button" class="vl-th-sortable" data-sort="gender">Gender <span class="mdi mdi-sort" aria-hidden="true"></span></button>
+              <button type="button" class="vl-th-sortable" data-sort="benchmark">Speed <span class="mdi mdi-sort" aria-hidden="true"></span></button>
+              <button type="button" class="vl-th-sortable" data-sort="rating">Rating <span class="mdi mdi-sort" aria-hidden="true"></span></button>
+            </div>
             <div id="voice-list" tabindex="0" role="region" aria-label="Voice library"></div>
             <div class="vl-toolbar">
-              <button class="btn-secondary vl-tb-btn" id="refresh-voices-btn" title="Refresh voice list"><span class="mdi mdi-refresh"></span> Refresh</button>
+              <button class="btn-secondary vl-tb-btn" id="refresh-voices-btn" title="Refresh voice list"><span class="mdi mdi-refresh" aria-hidden="true"></span><span class="vl-tb-label"> Refresh</span></button>
-              <button class="btn-secondary vl-tb-btn" id="sync-voice-folders-btn" title="Sync active/hidden folders">Sync</button>
+              <button class="btn-secondary vl-tb-btn" id="sync-voice-folders-btn" title="Sync active/hidden folders"><span class="mdi mdi-sync" aria-hidden="true"></span><span class="vl-tb-label"> Sync</span></button>
-              <button class="btn-secondary vl-tb-btn vl-tb-bundle" id="export-voices-btn" title="Export the checked voices as .voice.zip bundles (backup / share)"><span class="mdi mdi-export" aria-hidden="true"></span> Export selected</button>
+              <button class="btn-secondary vl-tb-btn" id="import-voices-btn" title="Import a .voice.zip bundle; an existing voice is never overwritten"><span class="mdi mdi-import" aria-hidden="true"></span><span class="vl-tb-label"> Import bundle</span></button>
-              <button class="btn-secondary vl-tb-btn vl-tb-bundle" id="import-voices-btn" title="Import a .voice.zip bundle; an existing voice is never overwritten"><span class="mdi mdi-import" aria-hidden="true"></span> Import bundle</button>
               <input type="file" id="import-voices-input" accept=".zip,application/zip" multiple hidden>
-              <button class="btn-secondary vl-tb-btn" id="calculate-db-btn" title="Calculate dBFS">Calc dB</button>
+              <div class="vl-actions">
-              <button class="btn-secondary vl-tb-btn" id="benchmark-voices-btn" title="Benchmark TTS speed">Benchmark</button>
+                <button type="button" class="btn-primary vl-actions-btn" id="vl-actions-btn" aria-expanded="false" aria-controls="vl-actions-menu" disabled title="Check one or more voices to enable actions"><span class="mdi mdi-lightning-bolt-outline" aria-hidden="true"></span> Actions <span class="vl-actions-n" id="vl-actions-n"></span> <span class="mdi mdi-chevron-up" aria-hidden="true"></span></button>
-              <button class="btn-secondary vl-tb-btn" id="copy-active-voices-btn" title="Copy selected voices (or all active if none checked)">Copy selected</button>
+                <div class="vl-actions-menu" id="vl-actions-menu" role="group" aria-label="Actions on the checked voices" hidden>
-              <button class="btn-secondary vl-tb-btn" id="precompute-embeddings-btn" title="Warm all active voices so the TTS engine pre-computes &amp; caches each speaker embedding (.pt) — makes first playback instant"><span class="mdi mdi-flash-outline"></span> Precompute</button>
+                  <div class="vl-menu-title">Edit</div>
-              <button class="btn-secondary vl-tb-btn" id="seed-batch-all-btn" title="Generate &amp; cache Seed Finder samples for every active voice (skips already-cached; resumable)"><span class="mdi mdi-dice-multiple-outline"></span> Batch seeds</button>
+                <button class="vl-menu-item" id="vl-bulk-tag" title="Add a tag to the selected voices, keeping their existing tags"><span class="mdi mdi-tag-outline" aria-hidden="true"></span> Set tag</button>
-              <button class="btn-secondary vl-tb-btn" id="verify-voices-stt-btn" title="Synthesize a test sentence with each voice, transcribe it back with Whisper, and compare to the original text — catches repeated/garbled/nonsense output that a duration-only benchmark misses"><span class="mdi mdi-check-decagram-outline"></span> Verify (STT)</button>
+                <button class="vl-menu-item" id="vl-bulk-source" title="Set source / origin on selected voices"><span class="mdi mdi-database-outline" aria-hidden="true"></span> Set source</button>
+                <button class="vl-menu-item" id="vl-bulk-rating" title="Set star rating on selected"><span class="mdi mdi-star-outline" aria-hidden="true"></span> Rate</button>
+                <button class="vl-menu-item" id="vl-bulk-hide" title="Disable (hide) selected voices"><span class="mdi mdi-eye-off-outline" aria-hidden="true"></span> Hide</button>
+                <button class="vl-menu-item" id="vl-bulk-unhide" title="Enable (unhide) selected voices"><span class="mdi mdi-eye-outline" aria-hidden="true"></span> Unhide</button>
+                  <div class="vl-menu-title">Audio</div>
+                <button class="vl-menu-item" id="precompute-embeddings-btn" title="Warm all active voices so the TTS engine pre-computes &amp; caches each speaker embedding (.pt) — makes first playback instant"><span class="mdi mdi-flash-outline"></span> Precompute</button>
+                <button class="vl-menu-item" id="calculate-db-btn" title="Calculate dBFS"><span class="mdi mdi-sine-wave" aria-hidden="true"></span> Calc dB</button>
+                <button class="vl-menu-item" id="benchmark-voices-btn" title="Benchmark TTS speed"><span class="mdi mdi-speedometer" aria-hidden="true"></span> Benchmark</button>
+                <button class="vl-menu-item" id="seed-batch-all-btn" title="Generate &amp; cache Seed Finder samples for every active voice (skips already-cached; resumable)"><span class="mdi mdi-dice-multiple-outline"></span> Batch seeds</button>
+                <button class="vl-menu-item" id="verify-voices-stt-btn" title="Synthesize a test sentence with each voice, transcribe it back with Whisper, and compare to the original text — catches repeated/garbled/nonsense output that a duration-only benchmark misses"><span class="mdi mdi-check-decagram-outline"></span> Verify (STT)</button>
+                  <div class="vl-menu-title">Share</div>
+                <button class="vl-menu-item" id="copy-active-voices-btn" title="Copy the names of the checked voices, comma separated (e.g. for SillyTavern)"><span class="mdi mdi-content-copy" aria-hidden="true"></span> Copy selected</button>
+                <button class="vl-menu-item" id="export-voices-btn" title="Export the checked voices as .voice.zip bundles (backup / share)"><span class="mdi mdi-export" aria-hidden="true"></span> Export selected</button>
+                  <div class="vl-menu-title">Danger</div>
+                <button class="vl-menu-item vl-bulk-danger" id="vl-bulk-delete" title="Delete selected voices permanently"><span class="mdi mdi-delete-outline" aria-hidden="true"></span> Delete</button>
+                </div>
+              </div>
             </div>
 
             <div class="vl-footer">

static/style.css

@@ -2104,7 +2104,13 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
     flex-shrink: 0;
   }
   .vl-card-list {
-    max-height: 55vh;
+    max-height: none;
+  }
+  /* The list scrolls inside a bounded box; the filter/synth panels and the toolbar
+     stay reachable by scrolling the page instead of squeezing the rows away. */
+  #voice-list {
+    flex: none;
+    max-height: max(50vh, 240px);
   }
   .voices-inspector-pane {
     flex: none;
@@ -2374,7 +2380,7 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
 .vl-sort-dir-btn:hover { background: var(--accent); color: #fff; border-color: var(--accent); }
 .vl-sort-dir-btn.active { background: var(--accent); color: #fff; border-color: var(--accent); }
 
-#voice-list { flex: 1; overflow-y: auto; min-height: 0; }
+#voice-list { flex: 1; overflow-y: auto; min-height: 160px; }
 
 /* Virtual tag folders in the voice list */
 .vl-tag-group { border-bottom: 1px solid var(--border); }
@@ -2411,6 +2417,7 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
 .vl-bulk-btn:hover { background: var(--panel); border-color: var(--accent); }
 .vl-bulk-danger { color: var(--red, #ef4444); border-color: rgba(239,68,68,.3); }
 .vl-bulk-danger:hover { background: rgba(239,68,68,.1); border-color: var(--red, #ef4444); }
+.vl-bulk-act:disabled { opacity: .45; cursor: not-allowed; }
 .vl-bulk-sep { width: 1px; height: 18px; background: var(--border); margin: 0 2px; flex-shrink: 0; }
 .vl-bulk-tag-inp, .vl-bulk-rating-sel {
   height: 26px; padding: 0 8px; border: 1px solid var(--accent); border-radius: 6px;
@@ -2483,6 +2490,14 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
   font-size: 15px; line-height: 1; padding: 0 2px; border-radius: 4px;
 }
 .vg-del:hover { color: var(--red, #ef4444); background: rgba(239,68,68,.12); }
+.vg-acc { flex: 1; min-width: 0; }
+.vg-acc > summary { cursor: pointer; list-style: none; display: flex; align-items: center; gap: 6px; }
+.vg-acc > summary::-webkit-details-marker { display: none; }
+.vg-acc > summary::after { content: "\F0140"; font: normal 16px/1 "Material Design Icons"; margin-left: auto; color: var(--subtext); }
+.vg-acc[open] > summary::after { content: "\F0143"; }
+.vg-acc > summary:focus-visible { outline: 2px solid var(--accent); outline-offset: 2px; border-radius: 4px; }
+.vg-active { font-weight: 600; color: #f59e0b; max-width: 55%; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.vg-chips { display: flex; flex-wrap: wrap; gap: 4px; padding-top: 6px; }
 .vg-clear { border: none; background: none; cursor: pointer; color: var(--accent); font-size: 12px; font-weight: 600; padding: 2px 8px; }
 .vl-footer {
   padding: 7px 10px; border-top: 1px solid var(--border);
@@ -2612,6 +2627,20 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
   position: sticky; top: 0; z-index: 5;
 }
 
+.vl-filter-icon { color: var(--subtext); font-size: 16px; }
+/* Compact sort header above the list in the narrow (list + inspector) layout; the
+   full table header takes over in table view. */
+.vl-compact-header {
+  display: flex; align-items: center; gap: 4px 10px; flex-wrap: wrap;
+  padding: 6px 10px; border-bottom: 1px solid var(--border); background: var(--surface);
+  font-size: 11px; font-weight: 700; text-transform: uppercase; letter-spacing: .05em; color: var(--subtext);
+}
+.vl-compact-header .vl-th-sortable { background: none; border: 0; padding: 2px 4px; font: inherit; color: inherit; text-transform: inherit; letter-spacing: inherit; }
+.vl-compact-header .vl-th-sortable:focus-visible { outline: 2px solid var(--accent); outline-offset: 1px; border-radius: 3px; }
+.vl-compact-header .vl-th-sortable.sort-asc, .vl-compact-header .vl-th-sortable.sort-desc { color: var(--accent); }
+.vl-compact-header .vl-th-sortable.sort-asc .mdi::before { content: "\F005D"; }
+.vl-compact-header .vl-th-sortable.sort-desc .mdi::before { content: "\F0045"; }
+.voices-workbench.table-view .vl-compact-header { display: none; }
 .vl-th-sortable { cursor: pointer; display: flex; align-items: center; gap: 3px; transition: color .15s; }
 .vl-th-sortable:hover { color: var(--text); }
 .vl-th-sortable .mdi { font-size: 12px; opacity: .6; }
@@ -3448,20 +3477,28 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
   padding: 0; width: min(460px, 94vw); box-shadow: 0 8px 32px rgba(0,0,0,.28);
   color: var(--text);
 }
+.dc-editing > :not(.add-engine-form) { display: none !important; }
+.dc-editing .add-engine-form { padding: 16px; min-width: 0; }
+.dc-editing .add-engine-field input,
+.dc-editing .add-engine-field select,
+.dc-editing .add-engine-field textarea { min-width: 0; width: 100%; box-sizing: border-box; }
+.dc-editing .add-engine-field input[type="checkbox"] { width: auto; }
+.dc-editing .settings-mini-actions { flex-wrap: wrap; }
 .add-engine-dlg::backdrop { background: rgba(0,0,0,.45); }
 .add-engine-form { padding: 22px 24px; display: flex; flex-direction: column; gap: 14px; }
 .add-engine-title { margin: 0 0 2px; font-size: 15px; font-weight: 700; }
 .add-engine-field { display: flex; flex-direction: column; gap: 5px; }
 .add-engine-field label { font-size: 12px; font-weight: 600; color: var(--subtext); }
 .add-engine-opt { font-weight: 400; opacity: .65; }
-.add-engine-field input, .add-engine-field select {
+.add-engine-field input, .add-engine-field select, .add-engine-field textarea {
   padding: 7px 10px; border: 1px solid var(--border); border-radius: 5px;
   background: var(--bg); color: var(--text); font-family: inherit; font-size: 13px;
   transition: border-color .15s;
 }
-.add-engine-field input:focus, .add-engine-field select:focus {
+.add-engine-field input:focus, .add-engine-field select:focus, .add-engine-field textarea:focus {
   outline: none; border-color: var(--accent);
 }
+.add-engine-field textarea { resize: vertical; min-height: 70px; }
 .add-engine-actions { display: flex; gap: 8px; justify-content: flex-end; padding-top: 4px; }
 /* Delete button variant */
 .dc-delete-btn { border-color: rgba(220,38,38,.35) !important; color: var(--red, #dc2626) !important; background: rgba(220,38,38,.05) !important; }
@@ -6064,3 +6101,64 @@ html[data-mag-night] #toast.success { color: var(--green); } html[data-mag-night
 .conv-chart details summary { cursor:pointer; min-height:44px; display:flex; align-items:center; font-size:12px; color:var(--subtext); }
 .conv-chart table { width:100%; font-size:12px; border-collapse:collapse; font-variant-numeric:tabular-nums; }
 .conv-chart th, .conv-chart td { text-align:left; padding:2px 4px; border-bottom:1px solid var(--border); }
+
+/* ── My Voices: top bar + collapsible filter/sort panel ───────────────── */
+.vl-topbar {
+  display: flex; align-items: center; gap: 6px; padding: 8px 10px;
+  border-bottom: 1px solid var(--border); background: var(--panel);
+}
+.vl-topbar .vl-search-main { flex: 1; min-width: 0; padding: 7px 10px; }
+.vl-filter-toggle { position: relative; }
+.vl-filter-badge {
+  position: absolute; top: -5px; right: -5px; min-width: 16px; height: 16px; padding: 0 4px;
+  border-radius: 8px; background: var(--accent); color: #fff; font-size: 10px; font-weight: 700;
+  line-height: 16px; text-align: center;
+}
+.vl-filter-panel { padding: 8px 10px 10px; border-bottom: 1px solid var(--border); background: var(--surface); }
+.vl-filter-panel[hidden] { display: none; }
+.vl-filter-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(130px, 1fr)); gap: 8px; }
+.vl-filter-grid label { display: flex; flex-direction: column; gap: 3px; font-size: 10px; font-weight: 700;
+  text-transform: uppercase; letter-spacing: .05em; color: var(--subtext); }
+.vl-filter-grid select { width: 100%; max-width: none; padding: 6px 8px; font-size: 13px; text-transform: none;
+  letter-spacing: 0; font-weight: 400; border: 1px solid var(--border); border-radius: var(--radius);
+  background: var(--bg); color: var(--text); }
+.vl-sort-dir-wide { width: 100%; height: 32px; }
+.vl-filter-actions { display: flex; align-items: center; justify-content: space-between; gap: 8px; flex-wrap: wrap; margin-top: 10px; }
+.vl-filter-panel .voice-groups-bar { margin: 10px -10px -10px; }
+/* Sort header: one line, columns share the width */
+.vl-compact-header { flex-wrap: nowrap; gap: 2px; }
+.vl-compact-header .vl-th-sortable { flex: 1 1 auto; justify-content: center; white-space: nowrap; }
+.vl-compact-header .vl-th-sortable:first-of-type { justify-content: flex-start; }
+.vl-filter-grid #voice-sort-field { width: 100%; padding: 6px 8px; font-size: 13px; }
+.vl-synth-drawer { border-bottom: 1px solid var(--border); background: var(--surface); }
+.vl-synth-drawer[hidden] { display: none; }
+
+/* ── My Voices bottom bar: utilities left, one Actions menu right ───────── */
+.vl-toolbar { align-items: center; }
+.vl-actions { margin-left: auto; position: relative; }
+.vl-actions-btn { display: inline-flex; align-items: center; gap: 5px; padding: 6px 12px; font-size: 13px; font-weight: 600; }
+.vl-actions-btn:disabled { opacity: .5; cursor: not-allowed; }
+.vl-actions-n { font-variant-numeric: tabular-nums; }
+.vl-actions-menu {
+  position: absolute; right: 0; bottom: calc(100% + 6px); z-index: 30; width: min(260px, 80vw);
+  max-height: min(80vh, 680px); overflow-y: auto; padding: 6px;
+  background: var(--surface); border: 1px solid var(--border); border-radius: var(--radius);
+  box-shadow: 0 8px 28px rgba(0,0,0,.35);
+}
+.vl-actions-menu[hidden] { display: none; }
+.vl-menu-title { padding: 6px 8px 2px; font-size: 10px; font-weight: 700; text-transform: uppercase;
+  letter-spacing: .06em; color: var(--subtext); }
+.vl-menu-item { display: flex; align-items: center; gap: 8px; width: 100%; min-height: 44px; padding: 7px 8px; border: 0;
+  border-radius: 6px; background: none; color: var(--text); font: inherit; font-size: 13px; text-align: left; cursor: pointer; }
+.vl-menu-item:hover, .vl-menu-item:focus-visible { background: var(--panel); outline: none; }
+.vl-menu-item:focus-visible { box-shadow: inset 0 0 0 2px var(--accent); }
+.vl-menu-item .mdi { font-size: 16px; color: var(--subtext); }
+.vl-menu-item.vl-bulk-danger, .vl-menu-item.vl-bulk-danger .mdi { color: var(--red, #ef4444); }
+.vl-menu-item.vl-bulk-danger:hover { background: rgba(239,68,68,.1); }
+/* One row: utilities collapse to icon buttons (label stays for screen readers); labels return when there is room. */
+.vl-toolbar { flex-wrap: nowrap; }
+.vl-toolbar .vl-tb-label { position: absolute; width: 1px; height: 1px; overflow: hidden; clip: rect(0,0,0,0); white-space: nowrap; }
+.vl-toolbar > .vl-tb-btn { min-width: 44px; min-height: 44px; display: inline-flex; align-items: center; justify-content: center; }
+.vl-toolbar > .vl-tb-btn .mdi { font-size: 18px; }
+@container (min-width: 520px) { .vl-toolbar .vl-tb-label { position: static; width: auto; height: auto; clip: auto; } }
+.vl-card-list { container-type: inline-size; }

tests/engine-connections.spec.js

@@ -0,0 +1,324 @@
+const { test, expect } = require('./_guard');
+const path = require('node:path');
+const { readFileSync } = require('node:fs');
+
+for (const [role, settingKey, modelKey] of [['llm', 'llm_url', 'llm_model'], ['ocr', 'vision_ocr_url', 'vision_ocr_model']]) {
+test(`${role} active panel shows persisted disconnect and restores models on reconnect`, async ({ page }) => {
+  let discoveries = 0;
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<!doctype html><html><body></body></html>' }));
+  await page.route('**/api/**', route => {
+    discoveries++;
+    return route.fulfill({ json: { ok: true, models: ['saved-model'] } });
+  });
+  await page.goto('/');
+  await page.setContent(`<select id="${role}-active-url"><option value="http://engine/v1">Engine</option></select><select id="${role}-active-model"></select><span id="${role}-active-status"></span><button id="${role}-active-refresh">Refresh</button>`);
+  await page.evaluate(({ settingKey, modelKey }) => {
+    window.$ = id => document.getElementById(id);
+    window.escHtml = value => String(value);
+    window._appSettings = { [settingKey]: 'http://engine/v1', [modelKey]: 'saved-model', engine_active: { [settingKey]: 'saved-engine' }, engine_disabled: { 'saved-engine': true } };
+  }, { settingKey, modelKey });
+  await page.addScriptTag({ path: path.resolve('static/js/ai-backends.js') });
+  await expect(page.locator(`#${role}-active-status`)).toHaveText('Disconnected');
+  expect(discoveries).toBe(0);
+  await page.evaluate(role => {
+    delete window._appSettings.engine_disabled['saved-engine'];
+    window[role === 'llm' ? 'syncActiveLlmPanel' : 'syncActiveOcrPanel']('http://engine/v1');
+  }, role);
+  await expect(page.locator(`#${role}-active-model`)).toHaveValue('saved-model');
+  await expect(page.locator(`#${role}-active-status`)).toHaveText('1 model');
+  expect(discoveries).toBe(1);
+  let delayedModels;
+  await page.route('**/api/engine-models?**', route => { delayedModels = route; });
+  await page.evaluate(role => window[role === 'llm' ? 'syncActiveLlmPanel' : 'syncActiveOcrPanel']('http://engine/v1'), role);
+  await expect.poll(() => !!delayedModels).toBe(true);
+  await expect(page.locator(`#${role}-active-refresh`)).toBeDisabled();
+  await page.evaluate(role => {
+    window._appSettings.engine_disabled['saved-engine'] = true;
+    window[role === 'llm' ? 'syncActiveLlmPanel' : 'syncActiveOcrPanel']('http://engine/v1');
+  }, role);
+  await delayedModels.fulfill({ json: { ok: true, models: ['stale-model'] } });
+  await expect(page.locator(`#${role}-active-refresh`)).toBeEnabled();
+  await expect(page.locator(`#${role}-active-status`)).toHaveText('Disconnected');
+  await expect(page.locator(`#${role}-active-model option`)).toHaveText('Disconnected');
+});
+}
+
+for (const [role, settingKey] of [['stt', 'faster_whisper_url'], ['tts', 'tts_url'], ['llm', 'llm_url'], ['ocr', 'vision_ocr_url']]) {
+test(`${role} cards test repeatedly and activate the chosen model with a saved key`, async ({ page }) => {
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<!doctype html><html><body></body></html>' }));
+  const activations = [];
+  const connections = [];
+  let rejectDisconnect = false;
+  const settings = { engine_local_urls: { 'dc-custom-stt': 'http://engine/v1' }, engine_models: { 'custom-stt': 'whisper-b' }, engine_disabled: {}, engine_api_keys: { 'custom-stt': '__configured__' } };
+  let probes = 0;
+  await page.route('**/api/**', async route => {
+    const url = new URL(route.request().url());
+    if (url.pathname === '/api/probe-url') {
+      probes++;
+      expect(url.searchParams.get('name')).toBe('custom-stt');
+      return route.fulfill({ json: { ok: true } });
+    }
+    if (url.pathname.endsWith('/connection')) {
+      const { connected } = route.request().postDataJSON();
+      if (!connected && rejectDisconnect) return route.fulfill({ status: 500, json: { ok: false, error: 'Save failed' } });
+      connections.push(connected);
+      if (connected) delete settings.engine_disabled['custom-stt']; else settings.engine_disabled['custom-stt'] = true;
+      return route.fulfill({ json: { ok: true, connected } });
+    }
+    if (url.pathname === '/api/engine-models') {
+      expect(url.searchParams.get('type')).toBe(role);
+      expect(url.searchParams.get('name')).toBe('custom-stt');
+      return route.fulfill({ json: { ok: true, models: ['whisper-a', 'whisper-b'] } });
+    }
+    if (/\/use-as-(engine|llm)$/.test(url.pathname)) {
+      expect(route.request().headers()['x-provider-api-key']).toBe('test-only-key');
+      activations.push(route.request().postDataJSON());
+      return route.fulfill({ json: { ok: true, key_set: true } });
+    }
+    return route.fulfill({ json: settings });
+  });
+  await page.goto('/');
+  await page.setContent(`<div data-page="${role}"><div id="dc-grid-${role}"></div></div>`);
+  await page.evaluate(() => {
+    window.$ = id => document.getElementById(id);
+    window._appSettings = { engine_api_keys: { 'custom-stt': '__configured__' } };
+    window.escHtml = value => String(value).replace(/[&<>"']/g, c => `&#${c.charCodeAt(0)};`);
+    window.toast = () => {};
+    window.initLlmSnippets = () => {};
+    window.engineApiKeyValue = () => '';
+    window.applyAndSaveSettings = async patch => Object.assign(window._appSettings, patch);
+    window.probeUrl = (url, type, apiKey, name) => fetch('/api/probe-url?' + new URLSearchParams({ url, type, name })).then(r => r.json());
+  });
+  await page.addScriptTag({ path: path.resolve('static/js/engines.js') });
+  await page.evaluate(role => renderLocalContainers([{ id: 'custom-stt', name: 'Display name', label: 'Test engine', role, isCustom: true, url: 'http://engine/v1' }]), role);
+  await expect(page.locator('.llm-local-name')).toHaveText('Test engine');
+  await expect(page.locator('.dc-apikey-inp')).toBeVisible();
+  await page.locator('.dc-connect-btn').click();
+  await page.locator('.dc-connect-btn').click();
+  expect(probes).toBe(2);
+  await expect(page.locator('.dc-use-btn')).not.toHaveClass(/active/);
+  await page.locator('.dc-models-btn').click();
+  await expect(page.locator('datalist option[value="whisper-b"]')).toHaveCount(1);
+  await page.locator('.dc-model-inp').fill('whisper-b');
+  await page.locator('.dc-model-inp').blur();
+  await page.locator('.dc-apikey-inp').fill('test-only-key');
+  await page.locator('.dc-use-btn').click();
+  await expect.poll(() => activations.length).toBe(1);
+  expect(activations[0]).toEqual(role === 'llm'
+    ? { url: 'http://engine/v1', model: 'whisper-b' }
+    : { url: 'http://engine/v1', role, setting_key: settingKey, model: 'whisper-b', activate: true });
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  rejectDisconnect = true;
+  await page.locator('.dc-disconnect-btn').click();
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  expect(settings.engine_disabled).toEqual({});
+  rejectDisconnect = false;
+  await page.locator('.dc-disconnect-btn').click();
+  await expect(page.locator('.dc-disconnect-btn')).toBeHidden();
+  await expect(page.locator('.dc-use-btn')).not.toHaveClass(/active/);
+  expect(settings.engine_disabled).toEqual({ 'custom-stt': true });
+  expect(settings.engine_models['custom-stt']).toBe('whisper-b');
+  expect(settings.engine_api_keys['custom-stt']).toBe('__configured__');
+  // A stale browser flag cannot reconnect an engine disabled in saved settings.
+  await page.evaluate(() => localStorage.setItem('dc-con-custom-stt', '1'));
+  await page.evaluate(role => renderLocalContainers([{ id: 'custom-stt', name: 'Renamed display label', role, isCustom: true, url: 'http://engine/v1' }]), role);
+  await expect(page.locator('.dc-disconnect-btn')).toBeHidden();
+  await expect(page.locator('.dc-model-inp')).toHaveValue('whisper-b');
+  await expect(page.locator('.dc-url-inp')).toHaveValue('http://engine/v1');
+  await page.locator('.dc-connect-btn').click();
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  expect(settings.engine_disabled).toEqual({});
+  expect(connections).toEqual([true, true, false, true]);
+  await page.route('**/api/probe-url?**', route => route.fulfill({ json: { ok: false, error: 'Denied' } }));
+  await page.locator('.dc-connect-btn').click();
+  await expect(page.locator('.dc-connect-btn')).not.toHaveClass(/\bok\b/);
+  await page.route('**/api/engine-models?**', route => route.fulfill({ json: { ok: false, error: 'Denied' } }));
+  await page.locator('.dc-models-btn').click();
+  await expect(page.locator('.dc-model-inp')).toHaveValue('whisper-b');
+  if (role === 'stt' || role === 'tts') {
+    const name = role === 'stt' ? 'nvidia-parakeet-asr' : 'fish-speech-api';
+    const setting = role === 'stt' ? 'nvidia_router_url' : 'fishspeech_url';
+    await page.evaluate(({ name, role }) => renderLocalContainers([{ name, role, port: 9000 }]), { name, role });
+    await page.locator('.dc-apikey-inp').fill('test-only-key');
+    await page.locator('.dc-model-inp').fill('actual-model-id');
+    await page.locator('.dc-use-btn').click();
+    await expect.poll(() => activations.length).toBe(2);
+    expect(activations[1]).toEqual({ url: 'http://host.docker.internal:9000', role, setting_key: setting, model: 'actual-model-id', activate: true });
+  }
+  if (role === 'tts') {
+    const targets = [
+      ['tts-faster-qwen3-tts-voiceclone', 'tts_url'],
+      ['tts-faster-qwen3-tts-voicedesign', 'voice_design_url'],
+      ['tts-faster-qwen3-tts-customvoice', 'customvoice_url'],
+      ['tts-faster-qwen3-tts-streaming', 'tts_stream_url'],
+      ['fish-speech-api', 'fishspeech_url'], ['kokoro-tts', 'kokoro_url'],
+      ['vibevoice-msft-tts', 'vibevoice_url'], ['nvidia-zeroshot-nim', 'nvidia_zeroshot_url'],
+      ['nvidia-flow-nim', 'nvidia_flow_url'], ['xtts-dgx-spark', 'xtts_url'],
+    ];
+    await page.evaluate(targets => renderLocalContainers(targets.map(([name]) => ({ name, role: 'tts', port: 9000 }))), targets);
+    expect((await page.locator('.dc-use-btn').allTextContents()).map(text => text.trim())).toEqual(targets.map(() => 'Use as TTS'));
+    expect(await page.locator('.dc-use-btn').evaluateAll(buttons => buttons.map(button => button.dataset.dcUseKey))).toEqual(targets.map(([, key]) => key));
+  }
+});
+}
+
+test('custom editor probes, saves credentials separately, edits inline by stable ID and keeps failed saves open', async ({ page }) => {
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<!doctype html><html><body></body></html>' }));
+  let settings = { custom_engine_cards: [], engine_models: {}, engine_api_keys: {} };
+  let rejectSave = false;
+  let delaySave = false;
+  let pendingSave;
+  const writes = [];
+  const checks = [];
+  const activations = [];
+  await page.route('**/api/**', async route => {
+    const request = route.request();
+    const url = new URL(request.url());
+    if (/\/use-as-(llm|engine)$/.test(url.pathname)) {
+      activations.push({ path: url.pathname, body: request.postDataJSON() });
+      return route.fulfill({ json: { ok: true } });
+    }
+    if (url.pathname === '/api/settings' && request.method() === 'POST') {
+      const body = request.postDataJSON();
+      writes.push(body);
+      if (delaySave) { pendingSave = route; return; }
+      if (rejectSave) return route.fulfill({ status: 500, json: { error: 'Save failed' } });
+      settings = { ...settings, ...body, engine_api_keys: { ...settings.engine_api_keys } };
+      for (const key of Object.keys(body.engine_api_keys || {})) settings.engine_api_keys[key] = '__configured__';
+      return route.fulfill({ json: { ok: true } });
+    }
+    if (['/api/probe-url', '/api/engine-models'].includes(url.pathname)) {
+      checks.push({ type: url.searchParams.get('type'), name: url.searchParams.get('name'), typed: request.headers()['x-provider-api-key'] });
+      return route.fulfill({ json: { ok: true, models: ['discovered-model'] } });
+    }
+    return route.fulfill({ json: settings });
+  });
+  await page.goto('/');
+  const section = readFileSync(path.resolve('static/sections/s-llms.html'), 'utf8');
+  await page.setContent(['stt', 'tts', 'llm', 'ocr'].map(role => `<div data-page="${role}"><div id="dc-grid-${role}"></div></div>`).join('') + section.slice(section.indexOf('<dialog'), section.indexOf('</dialog>') + 9));
+  await page.addStyleTag({ path: path.resolve('static/style.css') });
+  await page.evaluate(() => {
+    window.$ = id => document.getElementById(id);
+    window._appSettings = { custom_engine_cards: [], engine_models: {}, engine_api_keys: {} };
+    window.escHtml = value => String(value).replace(/[&<>"']/g, c => `&#${c.charCodeAt(0)};`);
+    window.toast = () => {};
+    window.initLlmSnippets = () => {};
+  });
+  await page.addScriptTag({ path: path.resolve('static/js/engines.js') });
+  await page.evaluate(() => openAddEngineDialog('stt'));
+  await page.locator('#aed-name').fill('Example STT');
+  await page.locator('#aed-url').fill('http://engine/v1');
+  await page.locator('#aed-api-key').fill('test-only-key');
+  for (const invalid of ['[1]', '{', '{"nested":{"api_key":"test-only-key"}}', '{"headers":{"x-api-key":"test-only-key"}}', '{"llm_api_key":"test-only-key"}', '{"BearerToken":"test-only-key"}']) {
+    await page.locator('#aed-params').fill(invalid);
+    await page.locator('#aed-save').click();
+    await page.locator('#aed-test').click();
+    expect(writes).toHaveLength(0);
+    expect(checks).toHaveLength(0);
+    await expect(page.locator('#add-engine-dlg')).toBeVisible();
+  }
+  await page.locator('#aed-params').fill('{"temperature":0.7,"max_tokens":50}');
+  await page.locator('#aed-active').check();
+  await page.locator('#aed-test').click();
+  await page.locator('#aed-load-models').click();
+  await expect(page.locator('#aed-models option')).toHaveAttribute('value', 'discovered-model');
+  await page.locator('#aed-model').fill('manual-model');
+  await page.locator('#aed-save').click();
+  await expect(page.locator('#add-engine-dlg')).not.toBeVisible();
+  expect(writes).toHaveLength(1);
+  const id = writes[0].custom_engine_cards[0].id;
+  expect(writes[0].engine_api_keys).toEqual({ [id]: 'test-only-key' });
+  expect(writes[0].engine_models[id]).toBe('manual-model');
+  expect(writes[0].custom_engine_cards[0].params).toEqual({ temperature: 0.7, max_tokens: 50 });
+  await expect.poll(() => activations.length).toBe(1);
+  expect(activations[0]).toEqual({ path: `/api/local-containers/${id}/use-as-engine`, body: { url: 'http://engine/v1', model: 'manual-model', role: 'stt', activate: true, setting_key: 'faster_whisper_url' } });
+  expect(checks).toEqual([{ type: 'stt', name: id, typed: 'test-only-key' }, { type: 'stt', name: id, typed: 'test-only-key' }]);
+  expect(JSON.stringify(writes[0].custom_engine_cards)).not.toContain('test-only-key');
+  expect(await page.evaluate(() => localStorage.getItem('engines-custom-cards'))).not.toContain('test-only-key');
+  await expect(page.locator('#dc-grid-stt .dc-edit-btn')).toBeVisible();
+  await page.locator('#dc-grid-stt .dc-edit-btn').click();
+  await expect(page.locator('#add-engine-dlg')).not.toBeVisible();
+  await expect(page.locator('#dc-grid-stt .add-engine-form')).toBeVisible();
+  await page.locator('#aed-name').fill('Cancelled change');
+  await page.locator('#aed-api-key').fill('discarded-test-key');
+  await page.locator('#aed-cancel').click();
+  await expect(page.locator('#dc-grid-stt .llm-local-name')).toHaveText('Example STT');
+  await expect(page.locator('#dc-grid-stt .dc-edit-btn')).toBeFocused();
+  await page.locator('#dc-grid-stt .dc-url-inp').fill('http://visible-override/v1');
+  await page.locator('#dc-grid-stt .dc-container-name-inp').fill('visible-container');
+  await page.locator('#dc-grid-stt .dc-edit-btn').click();
+  await expect(page.locator('#aed-url')).toHaveValue('http://visible-override/v1');
+  await expect(page.locator('#aed-container')).toHaveValue('visible-container');
+  await expect(page.locator('#aed-name')).toHaveValue('Example STT');
+  await expect(page.locator('#aed-api-key')).toHaveValue('');
+  await expect(page.locator('#aed-api-key')).toHaveAttribute('placeholder', /Configured/);
+  await expect(page.locator('#aed-model')).toHaveValue('manual-model');
+  await expect(page.locator('#aed-params')).toHaveValue(/"temperature": 0.7/);
+  await page.locator('#aed-active').uncheck();
+  await page.locator('#aed-name').fill('Renamed STT');
+  await page.locator('#aed-url').fill('http://changed-engine/v1');
+  let delayedModels;
+  await page.route('**/api/engine-models?**', route => { delayedModels = route; });
+  await page.locator('#aed-load-models').click();
+  await expect.poll(() => !!delayedModels).toBe(true);
+  await page.locator('#aed-url').fill('http://new-engine/v1');
+  await delayedModels.fulfill({ json: { ok: true, models: ['stale-model'] } });
+  await expect(page.locator('#aed-load-models')).toBeEnabled();
+  await expect(page.locator('#aed-models option')).toHaveCount(0);
+  delaySave = true;
+  await page.locator('#aed-save').click();
+  await expect.poll(() => !!pendingSave).toBe(true);
+  await expect(page.locator('#aed-cancel')).toBeDisabled();
+  await page.evaluate(() => {
+    openAddEngineDialog('tts');
+    editCustomEngineCard(window._appSettings.custom_engine_cards[0]);
+    renderLocalContainers([]);
+    loadLocalContainers();
+  });
+  await expect(page.locator('#aed-name')).toHaveValue('Renamed STT');
+  await expect(page.locator('#dc-grid-stt .add-engine-form')).toBeVisible();
+  delaySave = false;
+  await pendingSave.fulfill({ status: 500, json: { error: 'Save failed' } });
+  await expect(page.locator('#aed-save')).toBeEnabled();
+  await expect(page.locator('#aed-cancel')).toBeEnabled();
+  await expect(page.locator('#dc-grid-stt .add-engine-form')).toBeVisible();
+  await expect(page.locator('#aed-name')).toHaveValue('Renamed STT');
+  await expect(page.locator('#aed-url')).toHaveValue('http://new-engine/v1');
+  rejectSave = false;
+  await page.route('**/api/settings', route => route.request().method() === 'GET' ? route.abort() : route.fallback());
+  await page.locator('#aed-save').click();
+  await expect(page.locator('#add-engine-dlg')).not.toBeVisible();
+  await expect(page.locator('#dc-grid-stt .llm-local-name')).toHaveText('Renamed STT');
+  expect(writes.at(-1).custom_engine_cards[0].id).toBe(id);
+  expect(writes.at(-1)).not.toHaveProperty('engine_api_keys');
+  expect(writes.at(-1).engine_local_urls['dc-' + id]).toBe('http://new-engine/v1');
+  expect(await page.evaluate(id => localStorage.getItem('dc-url-' + id), id)).toBe('http://new-engine/v1');
+  await page.route('**/api/local-containers/*/use-as-llm', route => {
+    activations.push({ path: new URL(route.request().url()).pathname, body: route.request().postDataJSON() });
+    return route.fulfill({ status: 400, json: { ok: false, error: 'Denied' } });
+  });
+  await page.evaluate(() => {
+    window.engineTestMessages = [];
+    window.toast = message => window.engineTestMessages.push(message);
+    editCustomEngineCard(window._appSettings.custom_engine_cards[0]);
+  });
+  await page.locator('#aed-role').selectOption('llm');
+  await page.locator('#aed-active').check();
+  const savesBeforeActivation = writes.length;
+  await page.locator('#aed-save').click();
+  await expect(page.locator('#add-engine-dlg')).not.toBeVisible();
+  await expect.poll(() => page.evaluate(() => window.engineTestMessages.some(message => message.includes('saved; activation failed')))).toBe(true);
+  await expect(page.locator('#dc-grid-llm .llm-local-name')).toHaveText('Renamed STT');
+  await expect(page.locator('#dc-grid-stt .llm-local-card')).toHaveCount(0);
+  for (const role of ['ocr', 'tts', 'stt']) {
+    await page.evaluate(() => editCustomEngineCard(window._appSettings.custom_engine_cards[0]));
+    await expect(page.locator('.dc-editing .add-engine-form')).toBeVisible();
+    expect(await page.locator('.add-engine-form').evaluate(form => form.getBoundingClientRect().right <= innerWidth)).toBe(true);
+    await page.locator('#aed-role').selectOption(role);
+    await page.locator('#aed-active').uncheck();
+    await page.locator('#aed-save').click();
+    await expect(page.locator(`#dc-grid-${role} .dc-edit-btn`)).toBeVisible();
+  }
+  expect(writes).toHaveLength(savesBeforeActivation + 4);
+  expect(activations.at(-1)).toEqual({ path: `/api/local-containers/${id}/use-as-llm`, body: { url: 'http://new-engine/v1', model: 'manual-model' } });
+});

tests/engine-footer.spec.js

@@ -0,0 +1,312 @@
+const { test, expect } = require('./_guard');
+const path = require('node:path');
+
+test('footer chooses dynamic custom TTS ID and settings preserve custom STT preference', async ({ page }) => {
+  const errors = [];
+  page.on('pageerror', error => errors.push(String(error)));
+  let activated = false;
+  let rejectConnection = false;
+  let lastEngineOnly = false;
+  const connectionChanges = [];
+  const settings = {
+    custom_engine_cards: [
+      { id: 'stable-stt', name: 'Renamed STT', role: 'stt' },
+      { id: 'stable-tts', name: 'Renamed voice connection', role: 'tts', isCustom: true, url: 'http://engine/v1' },
+    ],
+    stt_preferred_backend: 'custom:stable-stt',
+    engine_disabled: {},
+    engine_models: { 'stable-tts': 'keep-model' },
+    engine_api_keys: { 'stable-tts': '__configured__' },
+    engine_local_urls: { 'dc-stable-tts': 'http://engine/v1' },
+  };
+  const backends = [
+    { id: 'openai', label: 'Voice Clone API', title: 'Qwen3 TTS VoiceClone', model: 'tts-1', available: true },
+    { id: 'custom:stable-tts', label: 'Different API label', title: 'Renamed voice connection', model: 'actual-inference-model', available: true },
+    { id: 'legacy', label: 'Legacy backend', model: 'legacy-model', available: true },
+    { id: 'custom:offline', label: 'Offline', available: false },
+  ];
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<!doctype html><html><body></body></html>' }));
+  await page.route('**/api/**', route => {
+    const url = new URL(route.request().url());
+    if (url.pathname.endsWith('/connection')) {
+      if (rejectConnection) return route.fulfill({ status: 400, json: { ok: false, error: 'Denied' } });
+      const connected = route.request().postDataJSON().connected;
+      connectionChanges.push(connected);
+      if (connected) delete settings.engine_disabled['stable-tts']; else settings.engine_disabled['stable-tts'] = true;
+      return route.fulfill({ json: { ok: true, connected } });
+    }
+    if (url.pathname === '/api/probe-url') return route.fulfill({ json: { ok: true } });
+    if (url.pathname.endsWith('/use-as-engine')) { activated = true; return route.fulfill({ json: { ok: true } }); }
+    if (url.pathname === '/api/tts-backends') {
+      const visible = activated ? backends.filter(backend => backend.id !== 'custom:stable-tts' || !settings.engine_disabled['stable-tts']) : [backends[0]];
+      return route.fulfill({ json: { backends: lastEngineOnly ? visible.filter(backend => backend.id === 'custom:stable-tts') : visible } });
+    }
+    return route.fulfill({ json: settings });
+  });
+  await page.goto('/');
+  await page.setContent('<button class="status-engine" data-status-kind="tts">TTS</button><select id="tts-backend-select"><option value="openai">Default</option></select><select id="s-stt-preferred-backend"><option value="">Configured</option></select><div data-page="tts"><div id="dc-grid-tts"></div></div>');
+  await page.evaluate(() => {
+    window._sttBackends = [];
+    window.logErr = () => {};
+    window.initLlmSnippets = () => {};
+  });
+  await page.addScriptTag({ path: path.resolve('static/js/utils.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/settings.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/engines.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/ai-backends.js') });
+  await page.evaluate(settings => {
+    window._appSettings = settings;
+    window._ttsBackends = [{ id: 'openai', label: 'Default', available: true }];
+    renderLocalContainers([settings.custom_engine_cards[1]]);
+  }, settings);
+  await page.locator('.dc-use-btn').click();
+  await expect(page.locator('#tts-backend-select')).toHaveValue('custom:stable-tts');
+  await expect(page.locator('#tts-backend-select option[value="openai"]')).toHaveText('Qwen3 TTS VoiceClone');
+  await expect(page.locator('#tts-backend-select option[value="custom:stable-tts"]')).toHaveText('Renamed voice connection');
+  await expect(page.locator('#tts-backend-select option[value="legacy"]')).toHaveText('Legacy backend');
+  expect(await page.evaluate(() => {
+    const select = document.createElement('select');
+    select.innerHTML = styleBackendOptions('custom:stable-tts');
+    return [...select.options].map(option => ({ id: option.value, label: option.textContent, selected: option.selected }));
+  })).toEqual([
+    { id: 'openai', label: 'Qwen3 TTS VoiceClone', selected: false },
+    { id: 'custom:stable-tts', label: 'Renamed voice connection', selected: true },
+    { id: 'legacy', label: 'Legacy backend', selected: false },
+  ]);
+  await page.locator('.status-engine').click();
+  expect(errors).toEqual([]);
+  await expect(page.locator('.status-flyup-item[data-val="custom:stable-tts"]')).toHaveText('Renamed voice connection');
+  await expect(page.locator('.status-flyup-item[data-val="openai"]')).toHaveText('Qwen3 TTS VoiceClone');
+  await expect(page.locator('.status-flyup-item[data-val="legacy"]')).toHaveText('Legacy backend');
+  await expect(page.locator('.status-flyup-item[data-val="custom:offline"]')).toHaveCount(0);
+  await page.locator('.status-flyup-item[data-val="custom:stable-tts"]').click();
+  await expect(page.locator('#tts-backend-select')).toHaveValue('custom:stable-tts');
+  await page.locator('#tts-backend-select').selectOption('openai');
+  await page.locator('.status-engine').click();
+  await expect(page.locator('.status-flyup-item[data-val="openai"]')).toHaveClass(/is-active/);
+  await page.locator('.status-flyup-item[data-val="custom:stable-tts"]').click();
+  await expect(page.locator('#tts-backend-select')).toHaveValue('custom:stable-tts');
+  await page.evaluate(() => {
+    syncPreferredSttBackend(window._appSettings);
+    syncPreferredSttBackend(window._appSettings);
+  });
+  await expect(page.locator('#s-stt-preferred-backend')).toHaveValue('custom:stable-stt');
+  await expect(page.locator('#s-stt-preferred-backend option[value="custom:stable-stt"]')).toHaveCount(1);
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  rejectConnection = true;
+  await page.locator('.dc-disconnect-btn').click();
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  await expect(page.locator('#tts-backend-select')).toHaveValue('custom:stable-tts');
+  rejectConnection = false;
+  await page.locator('.status-engine').click();
+  await page.locator('.dc-disconnect-btn').click();
+  await expect(page.locator('.status-flyup')).toHaveCount(0);
+  await expect(page.locator('#tts-backend-select option[value="custom:stable-tts"]')).toHaveCount(0);
+  await expect(page.locator('.dc-disconnect-btn')).toBeHidden();
+  await expect(page.locator('.dc-url-inp')).toHaveValue('http://engine/v1');
+  await expect(page.locator('.dc-model-inp')).toHaveValue('keep-model');
+  expect(await page.evaluate(() => window._appSettings.engine_api_keys['stable-tts'] === '__configured__')).toBe(true);
+  await page.evaluate(() => {
+    localStorage.setItem('dc-con-stable-tts', '1');
+    renderLocalContainers([window._appSettings.custom_engine_cards[1]]);
+  });
+  await expect(page.locator('.dc-disconnect-btn')).toBeHidden();
+  await expect(page.locator('.dc-connect-btn')).not.toHaveClass(/\bok\b/);
+  await page.locator('.dc-connect-btn').click();
+  await expect(page.locator('.dc-disconnect-btn')).toBeVisible();
+  await expect(page.locator('#tts-backend-select option[value="custom:stable-tts"]')).toHaveCount(1);
+  await expect(page.locator('#tts-backend-select')).toHaveValue('openai');
+  expect(connectionChanges).toEqual([false, true]);
+  lastEngineOnly = true;
+  await page.evaluate(() => refreshTtsBackendAvailability('custom:stable-tts'));
+  await page.locator('.dc-disconnect-btn').click();
+  await expect(page.locator('#tts-backend-select')).toBeDisabled();
+  await expect(page.locator('#tts-backend-select')).toHaveValue('');
+});
+
+test('LLM footer selection updates active, rehearser and conversation model pickers', async ({ page }) => {
+  const writes = [];
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<!doctype html><html><body></body></html>' }));
+  await page.route('**/api/**', route => {
+    if (route.request().method() === 'POST') writes.push(route.request().postDataJSON());
+    return route.fulfill({ json: { ok: true, models: ['old-model', 'picked-model'] } });
+  });
+  await page.goto('/');
+  const options = '<option value="old-model">old-model</option><option value="picked-model">picked-model</option>';
+  await page.setContent('<button class="status-engine" data-status-kind="llm">LLM</button><select id="llm-active-url"><option value="http://engine/v1">Engine</option></select><span id="llm-active-status"></span>' +
+    ['llm-active-model', 'reh-llm-model', 'conv-llm-model-select'].map(id => `<select id="${id}">${options}</select>`).join(''));
+  await page.evaluate(() => {
+    window._appSettings = { llm_url: 'http://engine/v1', llm_model: 'old-model' };
+    window.logErr = () => {};
+    window.applyAndSaveSettings = patch => fetch('/api/settings', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(patch) });
+  });
+  await page.addScriptTag({ path: path.resolve('static/js/utils.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/benchmark-voice-picker.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/ai-backends.js') });
+  await expect(page.locator('#llm-active-status')).toHaveText('2 models');
+  await page.locator('.status-engine').click();
+  await page.locator('.status-flyup-item[data-val="picked-model"]').click();
+  for (const id of ['llm-active-model', 'reh-llm-model', 'conv-llm-model-select']) {
+    await expect(page.locator('#' + id)).toHaveValue('picked-model');
+  }
+  await expect(page.locator('#llm-active-status')).toHaveText('Active model: picked-model');
+  await expect(page.locator('.bvp-root[data-bvp-for="llm-active-model"] .bvp-trigger-content')).toHaveText('picked-model');
+  expect(await page.evaluate(() => window._appSettings.llm_model)).toBe('picked-model');
+  await expect.poll(() => writes.some(patch => patch.llm_model === 'picked-model')).toBe(true);
+  await page.evaluate(() => {
+    window._appSettings.engine_active = { llm_url: 'chosen' };
+    window._appSettings.engine_disabled = { chosen: true };
+    document.body.insertAdjacentHTML('beforeend', '<div id="status-bar"></div>');
+    window.updateStatusBar();
+  });
+  await expect(page.locator('[data-status-kind="llm"] .status-engine-value')).toHaveText('Disconnected');
+  await page.locator('.status-engine[data-status-kind="llm"]').first().click();
+  await expect(page.locator('.status-flyup')).toContainText('LLM disconnected');
+  await expect(page.locator('.status-flyup-item')).toHaveCount(0);
+  expect(await page.evaluate(() => window._appSettings.llm_model)).toBe('picked-model');
+});
+
+
+test('disconnect refresh keeps sidebar navigation on Engines and Settings', async ({ page }) => {
+  const fs = require('node:fs');
+  const index = fs.readFileSync(path.resolve('static/index.html'), 'utf8');
+  const tabs = index.match(/<nav class="tabs"[\s\S]*?<\/nav>/)[0];
+  const sidebar = index.match(/<aside id="sidebar"[\s\S]*?<\/aside>/)[0];
+  const errors = [];
+  const changes = [];
+  let noBackends = false;
+  page.on('pageerror', error => errors.push(String(error)));
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<html><body></body></html>' }));
+  await page.route('**/api/**', route => {
+    const url = new URL(route.request().url());
+    if (url.pathname.endsWith('/connection')) {
+      changes.push(route.request().postDataJSON().connected);
+      return route.fulfill({ json: { ok: true, connected: false } });
+    }
+    if (url.pathname === '/api/tts-backends') return route.fulfill({ json: { backends: noBackends ? [] : [{ id: 'openai', available: true }] } });
+    return route.fulfill({ json: { engine_disabled: {} } });
+  });
+  await page.goto('/#s-llms');
+  await page.setContent(tabs + sidebar + '<main id="main-content"><section id="s-llms" class="page-section"></section><section id="s-settings" class="page-section"></section><section id="s-tryout" class="page-section"></section></main>');
+  await page.addScriptTag({ path: path.resolve('static/js/utils.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/settings.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/engines.js') });
+  await page.evaluate(() => {
+    window.loadLocalContainers = () => {};
+    window.initLlmSnippets = () => {};
+    document.body.insertAdjacentHTML('beforeend', ['tts', 'stt', 'ocr', 'llm'].map(role => '<div data-page="' + role + '"><div id="dc-grid-' + role + '"></div></div>').join(''));
+  });
+  await page.addScriptTag({ path: path.resolve('static/nav.js') });
+  for (const section of ['s-llms', 's-settings']) {
+    await page.evaluate(section => navTo(section), section);
+    for (const role of ['tts', 'stt', 'ocr', 'llm']) {
+      noBackends = role === 'llm';
+      await page.evaluate(role => {
+        localStorage.setItem('dc-con-mock-' + role, '1');
+        renderLocalContainers([{ id: 'mock-' + role, name: 'Mock ' + role, role, isCustom: true, url: 'http://engine/v1' }]);
+      }, role);
+      const disconnect = page.locator('#dc-grid-' + role + ' .dc-disconnect-btn');
+      await disconnect.click();
+      await expect(disconnect).toBeHidden();
+      await expect(page).toHaveURL(new RegExp('#' + section + '$'));
+      await expect(page.locator('#' + section)).toHaveClass(/is-active/);
+      await expect(page.locator('[data-nav-section="' + section + '"]').first()).toHaveClass(/active/);
+    }
+  }
+  await expect(page.locator('.tab[data-tab="library"]')).toHaveClass(/backend-unavailable/);
+  expect(changes).toEqual(Array(8).fill(false));
+  expect(errors).toEqual([]);
+});
+
+test('legacy tabs still leave an unavailable active backend', async ({ page }) => {
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<html><body></body></html>' }));
+  await page.route('**/api/tts-backends', route => route.fulfill({ json: { backends: [] } }));
+  await page.goto('/');
+  await page.setContent('<div class="tab active" data-tab="library" data-backend-required="voice_clone"></div><div class="tab" data-tab="settings"></div>');
+  await page.addScriptTag({ path: path.resolve('static/js/utils.js') });
+  await page.addScriptTag({ path: path.resolve('static/js/settings.js') });
+  await page.evaluate(() => refreshTtsBackendAvailability());
+  await expect(page.locator('.tab[data-tab="settings"]')).toHaveClass(/active/);
+});
+
+test('STT connection and activation refresh card titles and chosen footer backend', async ({ page }) => {
+  const errors = [];
+  const writes = [];
+  page.on('pageerror', error => errors.push(String(error)));
+  const settings = {
+    stt_preferred_backend: 'faster_whisper', engine_disabled: { 'stable-stt': true },
+    custom_engine_cards: [{ id: 'stable-stt', name: 'DGX Realtime STT', role: 'stt', isCustom: true, url: 'http://engine/v1' }],
+  };
+  const backends = [
+    { id: 'configured', engine_id: 'stable-stt', label: 'Configured alias', title: 'DGX Realtime STT', available: true, connected: true },
+    { id: 'faster_whisper', engine_id: 'stable-stt', label: 'Configured alias', title: 'DGX Realtime STT', available: true, connected: true, url: 'http://engine/v1' },
+    { id: 'custom:stable-stt', engine_id: 'stable-stt', label: 'Inference model', title: 'DGX Realtime STT', available: true, connected: true, url: 'http://engine/v1' },
+    { id: 'custom:other-stt', engine_id: 'other-stt', label: 'Inference model', title: 'DGX Realtime STT', available: true, connected: true, url: 'http://engine/v1' },
+    { id: 'native', engine_id: 'native-stt', title: 'Disconnected native', available: true, connected: false },
+    { id: 'legacy-stt', label: 'Legacy STT', available: true },
+    { id: 'offline-stt', title: 'Offline STT', available: false },
+  ];
+  await page.route(url => url.pathname === '/', route => route.fulfill({ contentType: 'text/html', body: '<html><body></body></html>' }));
+  await page.route('**/api/**', route => {
+    const url = new URL(route.request().url());
+    if (url.pathname === '/api/probe-url') return route.fulfill({ json: { ok: true } });
+    if (url.pathname.endsWith('/connection')) {
+      delete settings.engine_disabled['stable-stt'];
+      return route.fulfill({ json: { ok: true } });
+    }
+    if (url.pathname.endsWith('/use-as-engine')) {
+      settings.stt_preferred_backend = 'custom:stable-stt';
+      return route.fulfill({ json: { ok: true } });
+    }
+    if (url.pathname === '/api/stt-backends') return route.fulfill({ json: { backends: backends.filter(b => b.id !== 'custom:stable-stt' || !settings.engine_disabled['stable-stt']) } });
+    if (url.pathname === '/api/settings' && route.request().method() === 'POST') {
+      const patch = route.request().postDataJSON();
+      writes.push(patch);
+      Object.assign(settings, patch);
+    }
+    return route.fulfill({ json: settings });
+  });
+  await page.goto('/');
+  const ids = ['conv-stt-select', 'stt-tts-stt-backend', 'clone-stt-backend'];
+  await page.setContent('<button class="status-engine" data-status-kind="stt">STT</button><select id="s-stt-preferred-backend"><option value="faster_whisper">Configured</option><option value="legacy-stt">Legacy</option></select>' +
+    ids.map(id => `<select id="${id}"><option value="faster_whisper">Stale</option></select>`).join('') + '<div data-page="stt"><div id="dc-grid-stt"></div></div>');
+  await page.evaluate(settings => { window._appSettings = settings; window.logErr = () => {}; window.initLlmSnippets = () => {}; }, settings);
+  for (const file of ['utils', 'settings', 'stt', 'engines', 'ai-backends']) await page.addScriptTag({ path: path.resolve('static/js/' + file + '.js') });
+  await page.evaluate(settings => { window._appSettings = settings; }, settings);
+  await page.evaluate(() => refreshSttBackends());
+  await page.evaluate(() => renderLocalContainers(window._appSettings.custom_engine_cards));
+  await page.locator('.dc-connect-btn').click();
+  for (const id of ids) {
+    await expect(page.locator('#' + id)).toHaveValue('faster_whisper');
+    await expect(page.locator('#' + id + ' option[value="custom:stable-stt"]')).toHaveCount(0);
+  }
+  await page.locator('.dc-use-btn').click();
+  for (const id of ids) {
+    await expect(page.locator('#' + id)).toHaveValue('custom:stable-stt');
+    await expect(page.locator('#' + id + ' option[value="custom:stable-stt"]')).toHaveText('DGX Realtime STT');
+  }
+  await expect(page.locator('#conv-stt-select option[value="faster_whisper"]')).toHaveCount(0);
+  await expect(page.locator('#conv-stt-select option[value="configured"]')).toHaveCount(0);
+  await expect(page.locator('#conv-stt-select option[value="custom:other-stt"]')).toHaveCount(1);
+  await expect(page.locator('#conv-stt-select option[value="native"]')).toBeDisabled();
+  await page.locator('#conv-stt-select').selectOption('legacy-stt');
+  await page.locator('.status-engine').click();
+  await expect(page.locator('.status-flyup-item[data-val="custom:stable-stt"]')).toHaveClass(/is-active/);
+  await expect(page.locator('.status-flyup-item[data-val="faster_whisper"]')).toHaveCount(0);
+  await expect(page.locator('.status-flyup-item[data-val="configured"]')).toHaveCount(0);
+  await expect(page.locator('.status-flyup-item[data-val="custom:other-stt"]')).toHaveCount(1);
+  await expect(page.locator('.status-flyup-item[data-val="native"]')).toHaveCount(0);
+  await expect(page.locator('.status-flyup-item[data-val="legacy-stt"]')).toHaveText('Legacy STT');
+  await expect(page.locator('.status-flyup-item[data-val="offline-stt"]')).toHaveCount(0);
+  await page.locator('.status-flyup-item[data-val="legacy-stt"]').click();
+  for (const id of [...ids, 's-stt-preferred-backend']) await expect(page.locator('#' + id)).toHaveValue('legacy-stt');
+  await expect.poll(() => writes.some(patch => patch.stt_preferred_backend === 'legacy-stt')).toBe(true);
+  await page.evaluate(() => { document.body.insertAdjacentHTML('beforeend', '<div id="status-bar"></div>'); updateStatusBar(); });
+  await expect(page.locator('[data-status-kind="stt"] .status-engine-value')).toHaveText('Legacy STT');
+  expect(await page.evaluate(() => sttBackendChoices(_sttBackends, 'faster_whisper').filter(b => b.engine_id === 'stable-stt' && b.available).map(b => b.id))).toEqual(['faster_whisper']);
+  expect(await page.evaluate(() => sttBackendChoices([
+    { id: 'nvidia_parakeet', engine_id: 'nvidia-parakeet-asr', available: true },
+    { id: 'nvidia_router', engine_id: 'nvidia-parakeet-asr', available: true },
+  ], 'nvidia_router').map(b => b.id))).toEqual(['nvidia_router']);
+  expect(errors).toEqual([]);
+});

tests/test_custom_tts.py

@@ -0,0 +1,212 @@
+"""Custom backend selection keeps endpoint, model, and credentials together."""
+import asyncio
+import os
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import AsyncMock, Mock, patch
+from fastapi import HTTPException
+
+os.environ.setdefault("CONFIG_DIR", tempfile.mkdtemp(prefix="custom-tts-tests-"))
+
+from core.config import _preview_backend_base_url
+from core import tts_helpers
+from routes import docker, stt, tts
+
+
+class CustomTtsTests(unittest.TestCase):
+    def settings(self):
+        return {
+            "tts_url": "http://default/v1", "tts_api_key": "default-key", "tts_backend": "openai",
+            "llm_url": "http://gateway/v1", "llm_api_key": "gateway-key",
+            "custom_engine_cards": [
+                {"id": "first", "name": "Not a model", "role": "tts", "url": "http://gateway/v1", "params": {"speed": 0.8}},
+                {"id": "chosen", "name": "Also not a model", "role": "tts", "url": "http://gateway/v1", "params": {"speed": 1.2}},
+            ],
+            "engine_api_keys": {"first": "first-key", "chosen": "chosen-key"},
+            "engine_models": {"first": "first-model", "chosen": "real-model"},
+        }
+
+    def test_custom_preview_posts_selected_model_key_and_params(self):
+        settings = self.settings()
+        response = Mock(content=b"audio", headers={"content-type": "audio/wav"}, status_code=200)
+        with patch("core.config._load_settings", return_value=settings), \
+             patch("core.config.request_with_retries", return_value=response) as post:
+            result = tts_helpers._preview_request_audio("hello", "alloy", settings, backend="custom:chosen")
+        self.assertEqual(result, (b"audio", "audio/wav"))
+        self.assertEqual(post.call_args.args, ("POST", "http://gateway/v1/audio/speech"))
+        payload = post.call_args.kwargs["json"]
+        self.assertEqual((payload["model"], payload["voice"], payload["input"], payload["speed"]),
+                         ("real-model", "alloy", "hello", 1.2))
+        self.assertEqual(post.call_args.kwargs["headers"], {"Authorization": "Bearer chosen-key"})
+
+    def test_url_priority_and_model_fallback_never_use_card_name(self):
+        settings = self.settings()
+        settings["engine_local_urls"] = {"dc-chosen": "http://override/v1", "chosen": "http://bare"}
+        self.assertEqual(_preview_backend_base_url(settings, "custom:chosen"), "http://override/v1")
+        settings["engine_models"].pop("chosen")
+        endpoint, payload, headers = tts_helpers._tts_request_config("hello", "voice", settings, backend_override="custom:chosen")
+        self.assertEqual(endpoint, "http://override/v1/audio/speech")
+        self.assertEqual(payload["model"], "tts-1")
+        self.assertEqual(headers, {"Authorization": "Bearer chosen-key"})
+        self.assertEqual(_preview_backend_base_url(settings, "custom:unknown"), "")
+
+    def test_keyless_gateway_probe_activation_and_inference_agree(self):
+        settings = self.settings()
+        settings["engine_api_keys"].pop("chosen")
+        for name in ("chosen",):
+            self.assertEqual(docker._engine_api_key(settings, name, "tts", url="http://gateway/"), "gateway-key")
+            self.assertEqual(docker._engine_api_key(settings, name, "tts", url="http://unrelated"), "sk-dummy-key")
+        self.assertEqual(docker._engine_api_key({"llm_url": "http://gateway/v1", "llm_api_key": "gateway-key"}, "", "tts", url="http://gateway"), "gateway-key")
+        with patch("routes.docker._load_settings", return_value=settings), patch("routes.docker._save_settings") as save:
+            request = Mock(json=AsyncMock(return_value={"url": "http://gateway/v1", "role": "tts", "setting_key": "tts_url", "model": "real-model"}))
+            asyncio.run(docker.use_as_engine("chosen", request, ""))
+        activated = save.call_args.args[0]
+        _, _, headers = tts_helpers._tts_request_config("hello", "voice", activated, backend_override="custom:chosen")
+        self.assertEqual(headers, {"Authorization": "Bearer gateway-key"})
+        self.assertEqual(docker._engine_api_key(settings, "chosen", "tts", "typed", "http://gateway"), "typed")
+
+    def test_unswitched_cards_list_with_card_url_and_skip_invalid_urls(self):
+        settings = self.settings()
+        settings["custom_engine_cards"][1]["label"] = "My Speech Engine"
+        settings["custom_engine_cards"].append({"id": "bad", "role": "tts", "url": ""})
+        with patch("routes.tts._load_settings", return_value=settings), \
+             patch("routes.tts._fetch_backend_voices", return_value=["alloy"]), \
+             patch("routes.tts._backend_health_detail", return_value={"ok": True}):
+            result = asyncio.run(tts.tts_backends())
+        custom = {item["id"]: item for item in result["backends"] if item["id"].startswith("custom:")}
+        self.assertEqual(set(custom), {"custom:first", "custom:chosen"})
+        self.assertEqual(custom["custom:chosen"]["url"], "http://gateway/v1")
+        self.assertEqual(custom["custom:chosen"]["label"], "real-model (Custom)")
+        self.assertEqual(custom["custom:chosen"]["title"], "My Speech Engine")
+        self.assertEqual(custom["custom:first"]["title"], "Not a model")
+        native = {item["id"]: item for item in result["backends"] if not item["id"].startswith("custom:")}
+        self.assertEqual(native["voice_clone"]["title"], "Qwen3 TTS · Voice Clone")
+        self.assertEqual(native["voice_design"]["title"], "Qwen3 TTS · Voice Design")
+        self.assertEqual(native["customvoice"]["title"], "Qwen3 TTS · Custom Voice")
+        self.assertEqual(native["streaming"]["title"], "Qwen3 TTS · Streaming")
+        self.assertEqual(native["fishspeech"]["title"], "Fish-Speech API")
+        self.assertEqual(settings["engine_models"]["chosen"], "real-model")
+
+    def test_model_discovery_does_not_request_voice_ids(self):
+        response = Mock(status_code=200, json=Mock(return_value={"data": [{"id": "real-model"}]}))
+        with patch("routes.docker._load_settings", return_value=self.settings()), \
+             patch("routes.docker.requests.get", return_value=response) as get:
+            result = asyncio.run(docker.engine_models("http://gateway/v1", "tts", "chosen", ""))
+        self.assertEqual(result["models"], ["real-model"])
+        self.assertEqual(get.call_args.args[0], "http://gateway/v1/models")
+        self.assertEqual(get.call_args.kwargs["headers"]["Authorization"], "Bearer chosen-key")
+
+    def test_speech_proxy_forwards_custom_backend(self):
+        async def bounded(fn, *args):
+            return fn(*args)
+
+        async def generate(text, _language, _settings, fn):
+            return await fn(text)
+
+        request = Mock(json=AsyncMock(return_value={"input": "hello", "voice": "alloy", "app": "test"}))
+        with patch("routes.tts._load_settings", return_value=self.settings()), \
+             patch("routes.tts._resolve_tts_route", return_value=("alloy", {"backend": "custom:chosen"})), \
+             patch("core.presets._resolve_virtual_voice", return_value=None), \
+             patch("routes.tts._resolve_virtual_voice", return_value=None), \
+             patch("routes.tts._bounded_tts_call", side_effect=bounded), \
+             patch("routes.tts._generate_with_tags", side_effect=generate), \
+             patch("routes.tts._preview_request_audio", return_value=(b"audio", "audio/wav")) as preview, \
+             patch("routes.tts._apply_route_sounds", return_value=(b"audio", "audio/wav", [])), \
+             patch("routes.tts._prepare_proxy_audio", return_value=(b"audio", "audio/wav", "wav", None, False)), \
+             patch("routes.tts._routing_log_request"):
+            response = asyncio.run(tts.openai_speech_proxy(request))
+        self.assertEqual(response.status_code, 200)
+        self.assertEqual(preview.call_args.args[-1], "custom:chosen")
+
+    def test_custom_stt_params_keep_selected_identity_on_shared_gateway(self):
+        settings = self.settings()
+        for card in settings["custom_engine_cards"]:
+            card["role"] = "stt"
+            card["params"] = {"language": "de" if card["id"] == "first" else "en"}
+        response = Mock(ok=True, status_code=200, text="transcribed")
+        with tempfile.NamedTemporaryFile(suffix=".wav") as audio, \
+             patch("routes.stt.requests.post", return_value=response) as post:
+            stt._transcribe_audio(Path(audio.name), settings, "custom:chosen")
+        self.assertEqual(post.call_args.kwargs["data"]["language"], "en")
+        self.assertEqual(post.call_args.kwargs["headers"]["Authorization"], "Bearer chosen-key")
+
+    def test_legacy_card_name_maps_remain_usable_before_migration(self):
+        settings = self.settings()
+        name = settings["custom_engine_cards"][1]["name"]
+        settings["engine_api_keys"] = {name: "legacy-key", "unrelated": "wrong-key"}
+        settings["engine_models"] = {name: "legacy-model"}
+        settings["engine_local_urls"] = {"dc-" + name: "http://legacy/v1"}
+        endpoint, payload, headers = tts_helpers._tts_request_config("hello", "voice", settings, backend_override="custom:chosen")
+        self.assertEqual(endpoint, "http://legacy/v1/audio/speech")
+        self.assertEqual(payload["model"], "legacy-model")
+        self.assertEqual(payload["speed"], 1.2)
+        self.assertEqual(headers, {"Authorization": "Bearer legacy-key"})
+        for role in ("tts", "stt", "ocr", "llm"):
+            self.assertEqual(docker._engine_api_key(settings, "chosen", role, url="http://legacy/v1"), "legacy-key")
+            self.assertEqual(docker._engine_api_key(settings, "unknown", role, url="http://unknown"), "sk-dummy-key")
+        settings["engine_api_keys"]["chosen"] = "id-key"
+        settings["engine_models"]["chosen"] = "id-model"
+        _, payload, headers = tts_helpers._tts_request_config("hello", "voice", settings, backend_override="custom:chosen")
+        self.assertEqual(payload["model"], "id-model")
+        self.assertEqual(headers, {"Authorization": "Bearer id-key"})
+        settings["engine_models"]["chosen"] = ""
+        _, payload, _ = tts_helpers._tts_request_config("hello", "voice", settings, backend_override="custom:chosen")
+        self.assertEqual(payload["model"], "tts-1")
+
+    def test_disconnect_persists_without_erasing_config_and_reconnect_restores(self):
+        settings = self.settings()
+        original_keys = dict(settings["engine_api_keys"])
+        original_models = dict(settings["engine_models"])
+        pairs = (("tts-faster-qwen3-tts-voiceclone", "voice_clone"), ("fish-speech-api", "fishspeech"), ("chosen", "custom:chosen"))
+        with patch("routes.docker._load_settings", side_effect=lambda: settings), \
+             patch("routes.docker._save_settings") as save:
+            for card, backend in pairs:
+                request = Mock(json=AsyncMock(return_value={"connected": False}))
+                self.assertEqual(asyncio.run(docker.engine_connection(card, request)), {"ok": True, "connected": False})
+                settings = save.call_args.args[0]
+        self.assertEqual(settings["engine_api_keys"], original_keys)
+        self.assertEqual(settings["engine_models"], original_models)
+        self.assertEqual(settings["tts_url"], "http://default/v1")
+        with patch("routes.tts._load_settings", side_effect=lambda: settings), \
+             patch("routes.tts._fetch_backend_voices", return_value=["alloy"]) as voices, \
+             patch("routes.tts._backend_health_detail", return_value={"ok": True}) as health:
+            listed = {item["id"]: item for item in asyncio.run(tts.tts_backends())["backends"]}
+            self.assertEqual(health.call_count, 10)
+            for card, backend in pairs:
+                self.assertTrue(listed[backend]["disabled"])
+                self.assertFalse(listed[backend]["available"])
+                self.assertNotIn(backend, [call.args[1] for call in voices.call_args_list])
+            with patch("routes.docker._load_settings", side_effect=lambda: settings), \
+                 patch("routes.docker._save_settings") as save:
+                for card, backend in pairs:
+                    asyncio.run(docker.engine_connection(card, Mock(json=AsyncMock(return_value={"connected": True}))))
+                    settings = save.call_args.args[0]
+            listed = {item["id"]: item for item in asyncio.run(tts.tts_backends())["backends"]}
+            for card, backend in pairs:
+                self.assertFalse(listed[backend]["disabled"])
+                self.assertTrue(listed[backend]["available"])
+
+    def test_connection_rejects_unknown_identity_and_invalid_boolean(self):
+        with patch("routes.docker._load_settings", return_value=self.settings()), \
+             patch("routes.docker._save_settings") as save:
+            for body in ([], {}, {"connected": "false"}, {"connected": 1}):
+                with self.assertRaises(HTTPException) as caught:
+                    asyncio.run(docker.engine_connection("chosen", Mock(json=AsyncMock(return_value=body))))
+                self.assertEqual(caught.exception.status_code, 400)
+            with self.assertRaises(HTTPException) as caught:
+                asyncio.run(docker.engine_connection("unknown", Mock(json=AsyncMock(return_value={"connected": False}))))
+            self.assertEqual(caught.exception.status_code, 404)
+            save.assert_not_called()
+
+    def test_use_as_reconnects_without_changing_other_disabled_cards(self):
+        settings = self.settings()
+        settings["engine_disabled"] = {"chosen": True, "first": True}
+        request = Mock(json=AsyncMock(return_value={"role": "tts", "setting_key": "tts_url", "url": "http://gateway/v1", "model": "real-model"}))
+        with patch("routes.docker._load_settings", return_value=settings), patch("routes.docker._save_settings") as save:
+            asyncio.run(docker.use_as_engine("chosen", request, ""))
+        self.assertEqual(save.call_args.args[0]["engine_disabled"], {"first": True})
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_engine_connections.py

@@ -0,0 +1,159 @@
+"""Provider credentials and model selections reach mocked inference calls."""
+import asyncio
+import os
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import AsyncMock, Mock, patch
+
+os.environ.setdefault('CONFIG_DIR', tempfile.mkdtemp(prefix='ttvc-engine-tests-'))
+from fastapi import HTTPException
+from routes import docker, stt, settings as settings_routes
+from core import tts_helpers
+from core.provider_health import provider_key_for_url, provider_request_params, validate_provider_params
+
+
+class EngineConnectionTests(unittest.TestCase):
+    def test_extra_params_are_role_scoped_and_required_fields_protected(self):
+        settings = {'custom_engine_cards': [
+            {'id': 'chat', 'url': 'http://gateway/v1', 'role': 'llm', 'params': {'temperature': 0.7}},
+            {'id': 'speech', 'url': 'http://gateway/v1', 'role': 'tts', 'params': {'speed': 1.2}},
+        ]}
+        payload = {'model': 'chosen', 'messages': [], 'temperature': 0.1}
+        merged = provider_request_params('http://gateway/v1/chat/completions', payload, settings)
+        self.assertEqual(merged['temperature'], 0.7)
+        self.assertNotIn('speed', merged)
+        self.assertEqual(merged['model'], 'chosen')
+        settings['custom_engine_cards'].append({'id': 'other-chat', 'url': 'http://gateway/v1', 'role': 'llm', 'params': {'top_p': 0.1}})
+        self.assertEqual(provider_request_params('http://gateway/v1/chat/completions', payload, settings), payload)
+
+    def test_params_follow_saved_endpoint_override(self):
+        settings = {'custom_engine_cards': [{'id': 'card', 'url': 'http://old', 'role': 'tts', 'params': {'speed': 1.2}}], 'engine_local_urls': {'dc-card': 'http://new/v1'}}
+        self.assertEqual(provider_request_params('http://new/v1/audio/speech', {'input': 'hello'}, settings)['speed'], 1.2)
+        self.assertEqual(provider_request_params('http://old/v1/audio/speech', {'input': 'hello'}, settings), {'input': 'hello'})
+
+    @patch('routes.settings._save_settings')
+    def test_invalid_or_secret_extra_params_are_rejected_before_save(self, save):
+        invalid = [[], {'model': 'replace'}, {'headers': {'x-api-key': 'dummy'}}, {'oauth': {'client_secret': 'dummy'}}, {'llm_api_key': 'dummy'}, {'nested': [{'BearerToken': 'dummy'}]}]
+        for params in invalid:
+            request = Mock(json=AsyncMock(return_value={'custom_engine_cards': [{'id': 'card', 'params': params}]}))
+            with self.assertRaises(HTTPException) as caught:
+                asyncio.run(settings_routes.post_settings(request))
+            self.assertEqual(caught.exception.status_code, 400)
+        save.assert_not_called()
+        validate_provider_params({'max_tokens': 100, 'temperature': 0.3})
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_explicit_activation_selects_custom_stt_and_enables_ocr(self, load, save):
+        for role, setting_key in [('stt', 'faster_whisper_url'), ('ocr', 'vision_ocr_url'), ('tts', 'tts_url')]:
+            load.return_value = {'custom_engine_cards': [{'id': 'card', 'role': role}]}
+            request = Mock(json=AsyncMock(return_value={'url': 'http://engine', 'role': role, 'setting_key': setting_key, 'activate': True}))
+            asyncio.run(docker.use_as_engine('card', request, ''))
+            saved = save.call_args.args[0]
+            if role == 'stt':
+                self.assertEqual(saved['stt_preferred_backend'], 'custom:card')
+            elif role == 'ocr':
+                self.assertTrue(saved['vision_ocr_enabled'])
+            else:
+                self.assertEqual(saved['tts_backend'], 'openai')
+
+    @patch('routes.settings._save_settings')
+    @patch('routes.settings._load_settings')
+    def test_custom_rename_migrates_legacy_maps_without_exposing_key(self, load, save):
+        load.return_value = {'custom_engine_cards': [{'id': 'custom-1', 'name': 'Old'}], 'engine_api_keys': {'Old': 'dummy-secret'}, 'engine_models': {'Old': 'selected'}, 'engine_local_urls': {'dc-Old': 'http://engine'}}
+        request = Mock(json=AsyncMock(return_value={'custom_engine_cards': [{'id': 'custom-1', 'name': 'Renamed'}], 'engine_api_keys': {'custom-1': '__configured__'}}))
+        result = asyncio.run(settings_routes.post_settings(request))
+        saved = save.call_args.args[0]
+        self.assertEqual(saved['engine_api_keys']['custom-1'], 'dummy-secret')
+        self.assertEqual(saved['engine_models']['custom-1'], 'selected')
+        self.assertEqual(saved['engine_local_urls']['dc-custom-1'], 'http://engine')
+        self.assertNotIn('dummy-secret', str(result))
+
+    @patch('routes.settings._save_settings')
+    @patch('routes.settings._load_settings')
+    def test_custom_migration_preserves_id_values_and_normal_cards(self, load, save):
+        load.return_value = {'custom_engine_cards': [{'id': 'custom-1', 'name': 'Old'}], 'engine_api_keys': {'Old': 'legacy', 'custom-1': 'current', 'ollama': 'normal'}, 'engine_models': {'Old': 'legacy', 'custom-1': 'current'}}
+        request = Mock(json=AsyncMock(return_value={'custom_engine_cards': [{'id': 'custom-1', 'name': 'Renamed'}]}))
+        asyncio.run(settings_routes.post_settings(request))
+        saved = save.call_args.args[0]
+        self.assertEqual(saved['engine_api_keys']['custom-1'], 'current')
+        self.assertEqual(saved['engine_api_keys']['ollama'], 'normal')
+        self.assertEqual(saved['engine_models']['custom-1'], 'current')
+
+    def test_card_credentials_do_not_use_other_provider_key(self):
+        settings = {'llm_api_key': 'unrelated', 'engine_api_keys': {'card': 'saved'}}
+        self.assertEqual(docker._engine_api_key(settings, 'card', 'llm', 'typed'), 'typed')
+        self.assertEqual(docker._engine_api_key(settings, 'card', 'llm'), 'saved')
+        self.assertEqual(docker._engine_api_key(settings, 'empty', 'llm'), 'sk-dummy-key')
+        self.assertEqual(docker._engine_headers('typed', 'anthropic')['x-api-key'], 'typed')
+
+    @patch('routes.docker.requests.get')
+    @patch('routes.docker._load_settings')
+    def test_model_list_uses_saved_card_key_and_root_fallback(self, load, get):
+        load.return_value = {'engine_api_keys': {'card': 'saved'}, 'llm_api_key': 'other'}
+        get.side_effect = [Mock(status_code=404), Mock(status_code=200, json=Mock(return_value={'data': [{'id': 'selected'}]}))]
+        result = asyncio.run(docker.engine_models('http://engine/v1', 'stt', 'card', ''))
+        self.assertEqual(result['models'], ['selected'])
+        self.assertEqual(get.call_args.args[0], 'http://engine/models')
+        self.assertEqual(get.call_args.kwargs['headers']['Authorization'], 'Bearer saved')
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_activation_scopes_key_and_clears_previous_model(self, load, save):
+        load.return_value = {'whisper_api_key': 'default', 'stt_models': {'faster_whisper': 'old'}}
+        request = Mock(json=AsyncMock(return_value={'url': 'http://engine', 'role': 'stt', 'setting_key': 'faster_whisper_url', 'model': ''}))
+        asyncio.run(docker.use_as_engine('card', request, 'typed'))
+        saved = save.call_args.args[0]
+        self.assertEqual(saved['whisper_api_key'], 'default')
+        self.assertEqual(saved['engine_api_keys']['card'], 'typed')
+        self.assertEqual(saved['stt_models']['faster_whisper'], '')
+        self.assertEqual(stt._stt_backend_api_key(saved, 'faster_whisper'), 'typed')
+
+    def test_activation_rejects_non_object_and_cross_role_setting(self):
+        for body in ([], {'role': 'ocr', 'setting_key': 'tts_url', 'url': 'http://engine'}):
+            request = Mock(json=AsyncMock(return_value=body))
+            with self.assertRaises(HTTPException) as caught:
+                asyncio.run(docker.use_as_engine('card', request, ''))
+            self.assertEqual(caught.exception.status_code, 400)
+
+    def test_endpoint_key_resolution_handles_ui_prefix(self):
+        settings = {'engine_local_urls': {'dc-card': 'http://engine/v1'}, 'engine_api_keys': {'card': 'saved'}}
+        self.assertEqual(provider_key_for_url(settings, 'http://engine', 'other'), 'saved')
+        settings['engine_api_keys'] = {}
+        self.assertEqual(provider_key_for_url(settings, 'http://engine', 'other'), '')
+
+    @patch('routes.stt.requests.post')
+    def test_stt_inference_uses_custom_card_key_and_model(self, post):
+        settings = {'custom_engine_cards': [{'id': 'card', 'url': 'http://engine/v1'}], 'engine_api_keys': {'card': 'saved'}, 'engine_models': {'card': 'selected'}, 'custom_engine_cards': [{'id': 'card', 'url': 'http://engine/v1', 'role': 'stt', 'params': {'language': 'en'}}]}
+        post.return_value = Mock(status_code=200, json=Mock(return_value={'text': 'heard'}))
+        with tempfile.TemporaryDirectory() as directory:
+            audio = Path(directory) / 'audio.wav'
+            audio.write_bytes(b'test')
+            stt._transcribe_audio(audio, settings, 'custom:card')
+        self.assertEqual(post.call_args.kwargs['headers']['Authorization'], 'Bearer saved')
+        self.assertEqual(post.call_args.kwargs['data']['model'], 'selected')
+        self.assertEqual(post.call_args.kwargs['data']['language'], 'en')
+
+    def test_tts_model_key_and_v1_base_reach_request_config(self):
+        settings = {'tts_url': 'http://engine/v1', 'tts_models': {'tts_url': 'selected'}, 'engine_local_urls': {'dc-card': 'http://engine/v1'}, 'engine_api_keys': {'card': 'saved'}, 'custom_engine_cards': [{'id': 'card', 'url': 'http://engine/v1', 'role': 'tts', 'params': {'speed': 1.2}}]}
+        endpoint, payload, headers = tts_helpers._tts_request_config('hello', 'voice', settings)
+        self.assertEqual(endpoint, 'http://engine/v1/audio/speech')
+        self.assertEqual(payload['model'], 'selected')
+        self.assertEqual(payload['speed'], 1.2)
+        self.assertEqual(headers['Authorization'], 'Bearer saved')
+
+    def test_explicit_tts_key_override_is_preserved(self):
+        settings = {'tts_url': 'http://primary', 'tts_api_key': 'default'}
+        _, _, headers = tts_helpers._tts_request_config('hello', 'voice', settings, url_override='http://other', api_key_override='default')
+        self.assertEqual(headers['Authorization'], 'Bearer default')
+
+    def test_sibling_backend_does_not_receive_default_credentials(self):
+        settings = {'tts_url': 'http://primary', 'tts_api_key': 'primary-key', 'whisper_url': 'http://primary', 'whisper_api_key': 'primary-key', 'faster_whisper_url': 'http://sibling'}
+        _, _, headers = tts_helpers._tts_request_config('hello', 'voice', settings, url_override='http://sibling')
+        self.assertEqual(headers, {})
+        self.assertEqual(stt._stt_backend_api_key(settings, 'faster_whisper'), '')
+
+
+if __name__ == '__main__':
+    unittest.main()

tests/test_engine_disconnect.py

@@ -0,0 +1,101 @@
+"""Disconnect gates inference without destroying saved engine configuration."""
+import asyncio
+import os
+import tempfile
+import unittest
+from unittest.mock import AsyncMock, Mock, patch
+
+os.environ.setdefault('CONFIG_DIR', tempfile.mkdtemp(prefix='ttvc-disconnect-tests-'))
+
+from fastapi import HTTPException
+from core.provider_health import engine_disconnected, provider_request_params
+from routes import docker, stt, tts
+
+
+class EngineDisconnectTests(unittest.TestCase):
+    def test_legacy_native_and_cross_role_endpoint_matching(self):
+        self.assertTrue(engine_disconnected({'engine_disabled': {'ollama': True}}, 'http://localhost:11434/v1', 'llm'))
+        self.assertTrue(engine_disconnected({'engine_disabled': {'ollama': True}}, 'http://host.docker.internal:11434/v1', 'llm'))
+        settings = {'engine_disabled': {'litellm': True}, 'engine_local_urls': {'litellm': 'http://gateway'}}
+        self.assertFalse(engine_disconnected(settings, 'http://gateway', 'tts'))
+        self.assertFalse(engine_disconnected(settings, 'http://gateway', 'ocr'))
+
+    @patch('routes.tts._backend_health_detail', return_value={'ok': True})
+    @patch('routes.tts._fetch_backend_voices', return_value=[])
+    @patch('routes.tts._load_settings')
+    def test_native_alias_of_disabled_custom_tts_is_unavailable(self, load, voices, health):
+        load.return_value = {'tts_url': 'http://gateway', 'engine_active': {'tts_url': 'custom-card'}, 'engine_disabled': {'custom-card': True}, 'custom_engine_cards': [{'id': 'custom-card', 'role': 'tts', 'url': 'http://gateway'}]}
+        payload = asyncio.run(tts.tts_backends())
+        item = next(item for item in payload['backends'] if item['id'] == 'voice_clone')
+        self.assertFalse(item['available'])
+        self.assertTrue(item['disabled'])
+        self.assertNotIn('http://gateway', [call.args[0] for call in health.call_args_list])
+
+    def test_shared_endpoint_uses_selected_identity(self):
+        settings = {'llm_url': 'http://gateway/v1', 'engine_active': {'llm_url': 'enabled'}, 'engine_local_urls': {'disabled': 'http://gateway', 'enabled': 'http://gateway'}, 'engine_disabled': {'disabled': True}}
+        self.assertFalse(engine_disconnected(settings, 'http://gateway/v1', 'llm'))
+        provider_request_params('http://gateway/v1/chat/completions', {}, settings)
+        settings['engine_active']['llm_url'] = 'disabled'
+        with self.assertRaisesRegex(ValueError, 'disconnected'):
+            provider_request_params('http://gateway/v1/chat/completions', {}, settings)
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_disconnect_retains_saved_configuration(self, load, save):
+        load.return_value = {'llm_url': 'http://gateway/v1', 'llm_model': 'chosen', 'engine_api_keys': {'card': 'dummy'}, 'custom_engine_cards': [{'id': 'card', 'role': 'llm'}]}
+        request = Mock(json=AsyncMock(return_value={'connected': False}))
+        asyncio.run(docker.engine_connection('card', request))
+        saved = save.call_args.args[0]
+        self.assertEqual(saved['llm_url'], 'http://gateway/v1')
+        self.assertEqual(saved['llm_model'], 'chosen')
+        self.assertEqual(saved['engine_api_keys']['card'], 'dummy')
+        self.assertTrue(saved['engine_disabled']['card'])
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_current_dc_url_overrides_stale_bare_url(self, load, save):
+        load.return_value = {'llm_url': 'http://current/v1', 'engine_local_urls': {'dc-card': 'http://current', 'card': 'http://stale'}, 'custom_engine_cards': [{'id': 'card', 'role': 'llm'}], 'engine_disabled': {'card': True}}
+        self.assertTrue(engine_disconnected(load.return_value, 'http://current', 'llm'))
+        self.assertFalse(engine_disconnected(load.return_value, 'http://stale', 'llm'))
+        request = Mock(json=AsyncMock(return_value={'connected': False}))
+        asyncio.run(docker.engine_connection('card', request))
+        self.assertEqual(save.call_args.args[0]['engine_active']['llm_url'], 'card')
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_disconnect_records_native_default_selection(self, load, save):
+        load.return_value = {'llm_url': 'http://host.docker.internal:11434/v1'}
+        request = Mock(json=AsyncMock(return_value={'connected': False}))
+        asyncio.run(docker.engine_connection('ollama', request))
+        self.assertEqual(save.call_args.args[0]['engine_active']['llm_url'], 'ollama')
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_disconnect_records_unique_legacy_selection_for_both_roles(self, load, save):
+        for role, setting in [('llm', 'llm_url'), ('ocr', 'vision_ocr_url')]:
+            load.return_value = {setting: 'http://lan:14000/v1', 'engine_local_urls': {'dc-card': 'http://lan:14000'}, 'custom_engine_cards': [{'id': 'card', 'role': role}]}
+            request = Mock(json=AsyncMock(return_value={'connected': False}))
+            asyncio.run(docker.engine_connection('card', request))
+            self.assertEqual(save.call_args.args[0]['engine_active'][setting], 'card')
+
+    @patch('routes.docker._save_settings')
+    @patch('routes.docker._load_settings')
+    def test_disconnect_preserves_other_selection_on_shared_endpoint(self, load, save):
+        load.return_value = {'llm_url': 'http://gateway/v1', 'engine_active': {'llm_url': 'other'}, 'engine_local_urls': {'card': 'http://gateway', 'other': 'http://gateway'}, 'custom_engine_cards': [{'id': 'card', 'role': 'llm'}, {'id': 'other', 'role': 'llm'}]}
+        request = Mock(json=AsyncMock(return_value={'connected': False}))
+        asyncio.run(docker.engine_connection('card', request))
+        self.assertEqual(save.call_args.args[0]['engine_active']['llm_url'], 'other')
+
+    @patch('routes.stt._stt_backend_health')
+    @patch('routes.stt._load_settings')
+    def test_disabled_stt_skips_probe_and_rejects_inference(self, load, health):
+        load.return_value = {'engine_disabled': {'faster-whisper': True}, 'faster_whisper_url': 'http://whisper'}
+        health.return_value = {'ok': True, 'models': []}
+        payload = asyncio.run(stt.stt_backends())
+        item = next(item for item in payload['backends'] if item['id'] == 'faster_whisper')
+        self.assertFalse(item['available'])
+        self.assertFalse(item['connected'])
+        self.assertNotIn('http://whisper', [call.args[0] for call in health.call_args_list])
+        with self.assertRaises(HTTPException) as caught:
+            stt._transcribe_audio(None, load.return_value, 'faster_whisper')
+        self.assertEqual(caught.exception.status_code, 409)

tests/test_magazine_auth.py

@@ -0,0 +1,67 @@
+"""Vision credentials stay attached to their configured endpoint."""
+import os
+import pathlib
+import tempfile
+import unittest
+from unittest import mock
+
+os.environ.setdefault("CONFIG_DIR", tempfile.mkdtemp(prefix="ttvc-magazine-auth-test-"))
+
+from core.magazine import BLOCK_FIGURE, describe_visual_region, vision_ocr_page, vision_ocr_page_chain
+
+
+class VisionAuthTests(unittest.TestCase):
+    def test_transcription_and_description_use_only_matching_vision_key(self):
+        settings = {
+            "vision_ocr_url": "http://primary/",
+            "vision_ocr_api_key": "primary-test-key",
+            "vision_ocr_fallback_url": "http://fallback/proxy/v1/",
+            "vision_ocr_fallback_api_key": "fallback-test-key",
+            "llm_api_key": "unrelated-test-key",
+        }
+        response = mock.Mock()
+        response.json.return_value = {"choices": [{"message": {"content":
+            "text [10,10,110,60]A real sentence transcribed from this page."}}]}
+        for url, expected in (
+            ("http://primary/v1/", {"Authorization": "Bearer primary-test-key"}),
+            ("http://fallback/proxy/v1", {"Authorization": "Bearer fallback-test-key"}),
+            ("http://unconfigured/v1", {}),
+            ("http://primary/other/v1", {}),
+        ):
+            for describe in (False, True):
+                with self.subTest(url=url, describe=describe), \
+                     mock.patch("core.config._load_settings", return_value=settings), \
+                     mock.patch("core.magazine.request_with_retries", return_value=response) as request:
+                    if describe:
+                        result = describe_visual_region(pathlib.Path(__file__), BLOCK_FIGURE, url, "model")
+                    else:
+                        result = vision_ocr_page(pathlib.Path(__file__), url, "model")
+                    self.assertIsNotNone(result)
+                    self.assertEqual(request.call_args.kwargs["headers"], expected)
+
+    def test_empty_key_and_fallback_never_reuse_another_engine_key(self):
+        settings = {
+            "vision_ocr_url": "http://primary/v1",
+            "vision_ocr_api_key": "primary-test-key",
+            "vision_ocr_fallback_url": "http://fallback/v1",
+            "vision_ocr_fallback_api_key": "",
+            "llm_api_key": "unrelated-test-key",
+        }
+        failure = mock.Mock()
+        failure.json.return_value = {"choices": [{"message": {"content": ""}}]}
+        success = mock.Mock()
+        success.json.return_value = {"choices": [{"message": {"content":
+            "text [10,10,110,60]A real sentence transcribed from this page."}}]}
+        with mock.patch("core.config._load_settings", return_value=settings), \
+             mock.patch("core.magazine.request_with_retries", side_effect=[failure, success]) as request:
+            result = vision_ocr_page_chain(pathlib.Path(__file__), [
+                ("http://primary/v1", "model"), ("http://fallback/v1", "model"),
+            ])
+        self.assertIsNotNone(result)
+        self.assertEqual(request.call_args_list[0].kwargs["headers"],
+                         {"Authorization": "Bearer primary-test-key"})
+        self.assertEqual(request.call_args_list[1].kwargs["headers"], {})
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_stt_footer_metadata.py

@@ -0,0 +1,137 @@
+"""STT footer metadata uses card names and effective endpoints."""
+import asyncio
+import os
+import tempfile
+import unittest
+from unittest.mock import patch
+from fastapi import HTTPException
+
+os.environ.setdefault('CONFIG_DIR', tempfile.mkdtemp(prefix='ttvc-stt-footer-tests-'))
+from routes import stt
+
+
+class SttFooterMetadataTests(unittest.TestCase):
+    def test_custom_models_use_own_legacy_name_with_stable_id_priority(self):
+        settings = {
+            'custom_engine_cards': [
+                {'id': 'first', 'name': 'First speech', 'url': 'http://shared', 'role': 'stt'},
+                {'id': 'second', 'name': 'Second speech', 'url': 'http://shared', 'role': 'stt'},
+            ],
+            'engine_models': {'First speech': 'first-legacy', 'Second speech': 'second-legacy'},
+        }
+        self.assertEqual(stt._stt_backend_model('custom:first', settings), 'first-legacy')
+        self.assertEqual(stt._stt_backend_model('custom:second', settings), 'second-legacy')
+        settings['engine_models']['first'] = 'first-current'
+        self.assertEqual(stt._stt_backend_model('custom:first', settings), 'first-current')
+        settings['engine_models']['first'] = ''
+        self.assertEqual(stt._stt_backend_model('custom:first', settings), 'large-v3')
+
+    def test_native_selected_credentials_preserve_only_matching_global_fallback(self):
+        settings = {
+            'engine_active': {'faster_whisper_url': 'faster-whisper'},
+            'faster_whisper_url': 'http://gateway/v1/',
+            'whisper_url': 'http://gateway',
+            'whisper_api_key': 'global-test-key',
+            'engine_api_keys': {'faster-whisper': 'own-test-key', 'other': 'other-test-key'},
+            'custom_engine_cards': [{'id': 'other', 'url': 'http://gateway', 'role': 'stt'}],
+        }
+        self.assertEqual(stt._stt_backend_api_key(settings, 'faster_whisper'), 'own-test-key')
+        del settings['engine_api_keys']['faster-whisper']
+        self.assertEqual(stt._stt_backend_api_key(settings, 'faster_whisper'), 'global-test-key')
+        settings['faster_whisper_url'] = 'http://unrelated'
+        self.assertEqual(stt._stt_backend_api_key(settings, 'faster_whisper'), '')
+        settings['engine_active']['faster_whisper_url'] = 'other'
+        settings['faster_whisper_url'] = 'http://gateway'
+        del settings['engine_api_keys']['other']
+        self.assertEqual(stt._stt_backend_api_key(settings, 'faster_whisper'), '')
+
+    @patch('routes.stt._stt_backend_health', return_value={'ok': True, 'models': ['discovered']})
+    @patch('routes.stt._load_settings')
+    def test_disabled_native_does_not_disable_selected_custom_on_shared_url(self, load, health):
+        load.return_value = {
+            'whisper_url': 'http://shared/v1',
+            'nvidia_router_url': 'http://shared/v1',
+            'engine_active': {'whisper_url': 'realtime', 'nvidia_router_url': 'realtime'},
+            'engine_disabled': {'nvidia-parakeet-asr': True},
+            'custom_engine_cards': [{'id': 'realtime', 'name': 'Realtime', 'role': 'stt', 'url': 'http://shared/v1'}],
+        }
+        items = {item['id']: item for item in asyncio.run(stt.stt_backends())['backends']}
+        self.assertFalse(items['nvidia_router']['available'])
+        self.assertFalse(items['nvidia_router']['connected'])
+        self.assertTrue(items['custom:realtime']['available'])
+        self.assertTrue(items['custom:realtime']['connected'])
+        self.assertEqual(items['configured']['engine_id'], 'realtime')
+        self.assertEqual(items['custom:realtime']['engine_id'], 'realtime')
+        self.assertEqual(sum(call.args[0] == 'http://shared/v1' for call in health.call_args_list), 2)
+        with self.assertRaises(HTTPException) as caught:
+            stt._transcribe_audio(None, load.return_value, 'nvidia_router')
+        self.assertEqual(caught.exception.status_code, 409)
+
+    @patch('routes.stt._stt_backend_health', return_value={'ok': True, 'models': ['discovered']})
+    @patch('routes.stt._load_settings')
+    def test_legacy_native_slot_uses_known_endpoint_identity(self, load, health):
+        load.return_value = {
+            'nvidia_asr_url': 'http://native:8090',
+            'engine_disabled': {'nvidia-parakeet-asr': True},
+        }
+        items = {item['id']: item for item in asyncio.run(stt.stt_backends())['backends']}
+        item = items['nvidia_parakeet']
+        self.assertEqual(item['title'], 'NVIDIA Parakeet Router ASR')
+        self.assertEqual(item['engine_id'], 'nvidia-parakeet-asr')
+        self.assertFalse(item['available'])
+        self.assertFalse(item['connected'])
+        self.assertNotIn('http://native:8090', [call.args[0] for call in health.call_args_list])
+        with self.assertRaises(HTTPException) as caught:
+            stt._transcribe_audio(None, load.return_value, 'nvidia_parakeet')
+        self.assertEqual(caught.exception.status_code, 409)
+
+    @patch('routes.stt._stt_backend_health', return_value={'ok': True, 'models': ['discovered']})
+    @patch('routes.stt._load_settings')
+    def test_selected_alternate_native_uses_its_own_disconnect_state(self, load, health):
+        load.return_value = {
+            'nvidia_asr_url': 'http://rnnt:8092',
+            'engine_active': {'nvidia_asr_url': 'nvidia-parakeet-rnnt-nim'},
+            'engine_disabled': {'parakeet-stt-multi-9011': True},
+        }
+        items = {item['id']: item for item in asyncio.run(stt.stt_backends())['backends']}
+        self.assertTrue(items['nvidia_parakeet']['available'])
+        self.assertTrue(items['nvidia_parakeet']['connected'])
+        self.assertEqual(items['nvidia_parakeet']['engine_id'], 'nvidia-parakeet-rnnt-nim')
+        self.assertFalse(stt._stt_disconnected(load.return_value, 'nvidia_parakeet', 'http://rnnt:8092'))
+        load.return_value['engine_disabled']['nvidia-parakeet-rnnt-nim'] = True
+        self.assertTrue(stt._stt_disconnected(load.return_value, 'nvidia_parakeet', 'http://rnnt:8092'))
+
+    @patch('routes.stt._stt_backend_health', return_value={'ok': True, 'models': ['discovered']})
+    @patch('routes.stt._load_settings')
+    def test_card_titles_stable_identity_and_effective_endpoint(self, load, health):
+        load.return_value = {
+            'faster_whisper_url': 'http://native:9000/v1',
+            'engine_active': {'faster_whisper_url': 'stt-whisperx-blackwell'},
+            'custom_engine_cards': [
+                {'id': 'first', 'name': 'Renamed speech', 'url': 'http://stale', 'role': 'stt'},
+                {'id': 'second', 'label': 'Other speech', 'url': 'http://shared/v1', 'role': 'stt'},
+                {'id': 'off', 'name': 'Disconnected', 'url': 'http://off', 'role': 'stt'},
+            ],
+            'engine_local_urls': {'first': 'http://old', 'dc-first': 'http://shared/v1'},
+            'engine_api_keys': {'first': 'first-test-key', 'second': 'second-test-key'},
+            'engine_models': {'first': 'first-model', 'second': 'second-model'},
+            'engine_disabled': {'off': True},
+        }
+        items = {item['id']: item for item in asyncio.run(stt.stt_backends())['backends']}
+        self.assertEqual(items['faster_whisper']['title'], 'WhisperX · faster-whisper (Blackwell GPU)')
+        self.assertEqual(items['custom:first']['title'], 'Renamed speech')
+        self.assertEqual(items['custom:second']['title'], 'Other speech')
+        self.assertEqual(items['custom:first']['url'], 'http://shared/v1')
+        self.assertEqual(stt._stt_backend_url(load.return_value, 'custom:first'), 'http://shared/v1')
+        self.assertEqual(items['custom:first']['model'], 'first-model')
+        self.assertEqual(items['custom:second']['model'], 'second-model')
+        self.assertTrue(items['custom:first']['available'])
+        health.assert_any_call('http://shared/v1', 'first-test-key')
+        health.assert_any_call('http://shared/v1', 'second-test-key')
+        self.assertFalse(items['custom:off']['available'])
+        self.assertFalse(items['custom:off']['connected'])
+        self.assertNotIn('http://off', [call.args[0] for call in health.call_args_list])
+
+
+if __name__ == '__main__':
+    unittest.main()

tests/test_tts_health_probe.py

@@ -0,0 +1,99 @@
+"""TTS reachability probing must work for OpenAI-style card URLs that end in /v1."""
+import os
+import tempfile
+import unittest
+from unittest.mock import patch
+
+os.environ.setdefault("CONFIG_DIR", tempfile.mkdtemp(prefix="tts-health-tests-"))
+
+from routes import tts
+
+
+class TtsHealthProbeTests(unittest.TestCase):
+    def test_v1_card_url_is_probed_from_the_server_root(self):
+        with patch("routes.tts.probe_provider", return_value={"ok": True}) as probe:
+            tts._backend_health_detail("http://gateway:4000/v1/", "key")
+        base, checks = probe.call_args.args[:2]
+        self.assertEqual(base, "http://gateway:4000")
+        self.assertIn(("/health/liveliness", None), checks)
+        self.assertEqual(probe.call_args.kwargs["headers"], {"Authorization": "Bearer key"})
+
+    def test_plain_host_url_is_unchanged(self):
+        with patch("routes.tts.probe_provider", return_value={"ok": False}) as probe:
+            tts._backend_health_detail("http://host:8005")
+        self.assertEqual(probe.call_args.args[0], "http://host:8005")
+
+
+
+class PublishUnhiddenVoiceTests(unittest.TestCase):
+    def setUp(self):
+        from pathlib import Path
+        self.dir = Path(tempfile.mkdtemp(prefix="publish-voice-"))
+        self.voices = self.dir / "voices.json"
+        self.settings = {"voices_scan_dir": str(self.dir), "tts_config_dir": str(self.dir)}
+        self.audio = self.dir / "DE_F_New.wav"
+        self.audio.write_bytes(b"RIFF")
+
+    def run_publish(self):
+        return tts.publish_voice_clone_if_missing(self.audio, "DE", self.settings)
+
+    def test_unknown_voice_with_transcript_is_published(self):
+        import json
+        self.voices.write_text(json.dumps({"other": {"ref_text": "x"}}))
+        self.audio.with_suffix(".reference.txt").write_text("Hallo Welt")
+        with patch("routes.tts.time.sleep"):
+            self.assertTrue(self.run_publish())
+        entry = json.loads(self.voices.read_text())["DE_F_New"]
+        self.assertEqual((entry["ref_text"], entry["language"]), ("Hallo Welt", "German"))
+
+    def test_known_voice_keeps_its_cached_embedding(self):
+        import json
+        self.voices.write_text(json.dumps({"DE_F_New": {"ref_text": "a", "speaker_embeddings": "/x.pt"}}))
+        self.audio.with_suffix(".reference.txt").write_text("Hallo")
+        self.assertFalse(self.run_publish())
+        self.assertIn("speaker_embeddings", json.loads(self.voices.read_text())["DE_F_New"])
+
+    def test_no_transcript_or_no_voices_file_is_a_no_op(self):
+        self.assertFalse(self.run_publish())  # no voices.json
+        self.voices.write_text("{}")
+        self.assertFalse(self.run_publish())  # no transcript
+        self.assertEqual(self.voices.read_text(), "{}")
+
+
+class GatewayUpstreamProbeTests(unittest.TestCase):
+    URL = "http://gateway:4000/v1"
+
+    def test_running_upstream_is_available(self):
+        with patch("routes.tts._gateway_upstream", return_value="http://host:8023/v1"), \
+                patch("routes.tts.probe_provider", return_value={"ok": True}) as probe:
+            self.assertTrue(tts._backend_health_detail(self.URL, "k", "m")["ok"])
+        self.assertEqual(probe.call_count, 1)
+        self.assertEqual(probe.call_args.args[0], "http://host:8023")
+
+    def test_stopped_upstream_is_not_rescued_by_a_live_gateway(self):
+        refused = {"ok": False, "error": "Connection refused"}
+        with patch("routes.tts._gateway_upstream", return_value="http://host:8020/v1"), \
+                patch("routes.tts.probe_provider", return_value=refused) as probe:
+            self.assertFalse(tts._backend_health_detail(self.URL, "k", "m")["ok"])
+        self.assertEqual(probe.call_count, 1)  # the gateway's /health/liveliness is not consulted
+
+    def test_unroutable_upstream_falls_back_to_the_gateway(self):
+        results = [{"ok": False, "error": "Timeout (2 s)"}, {"ok": True, "endpoint": "/health/liveliness"}]
+        with patch("routes.tts._gateway_upstream", return_value="http://internal:8020/v1"), \
+                patch("routes.tts.probe_provider", side_effect=results) as probe:
+            self.assertTrue(tts._backend_health_detail(self.URL, "k", "m")["ok"])
+        self.assertEqual(probe.call_args_list[1].args[0], "http://gateway:4000")
+
+    def test_upstream_lookup_needs_key_and_model_and_reads_model_info(self):
+        tts._GATEWAY_UPSTREAMS.clear()
+        self.assertEqual(tts._gateway_upstream("http://gw", "", "m"), "")
+        info = {"data": [{"model_name": "m", "litellm_params": {"api_base": "http://h:1/v1"}}, {"model_name": "x"}]}
+        resp = type("R", (), {"status_code": 200, "json": lambda self: info})()
+        with patch("routes.tts.requests.get", return_value=resp) as get:
+            self.assertEqual(tts._gateway_upstream("http://gw", "k", "m"), "http://h:1/v1")
+            self.assertEqual(tts._gateway_upstream("http://gw", "k", "other"), "")
+        self.assertEqual(get.call_count, 1)  # second lookup served from the cache
+
+
+if __name__ == "__main__":
+    unittest.main()